Commit ·
76862c3
0
Parent(s):
Release contextual text and image embedding model
Browse files- .gitattributes +36 -0
- README.md +175 -0
- assets/logo.svg +593 -0
- config.json +120 -0
- configuration_pplx_contextual.py +27 -0
- contextual_head.safetensors +3 -0
- image_processing.py +146 -0
- model-00001-of-00007.safetensors +3 -0
- model-00002-of-00007.safetensors +3 -0
- model-00003-of-00007.safetensors +3 -0
- model-00004-of-00007.safetensors +3 -0
- model-00005-of-00007.safetensors +3 -0
- model-00006-of-00007.safetensors +3 -0
- model-00007-of-00007.safetensors +3 -0
- model.safetensors.index.json +767 -0
- modeling_pplx_contextual.py +307 -0
- processor_config.json +63 -0
- tokenizer.json +3 -0
- tokenizer_config.json +36 -0
.gitattributes
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
*.7z filter=lfs diff=lfs merge=lfs -text
|
| 2 |
+
*.arrow filter=lfs diff=lfs merge=lfs -text
|
| 3 |
+
*.bin filter=lfs diff=lfs merge=lfs -text
|
| 4 |
+
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
| 5 |
+
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
| 6 |
+
*.ftz filter=lfs diff=lfs merge=lfs -text
|
| 7 |
+
*.gz filter=lfs diff=lfs merge=lfs -text
|
| 8 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
| 9 |
+
*.joblib filter=lfs diff=lfs merge=lfs -text
|
| 10 |
+
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
| 11 |
+
*.mlmodel filter=lfs diff=lfs merge=lfs -text
|
| 12 |
+
*.model filter=lfs diff=lfs merge=lfs -text
|
| 13 |
+
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
| 14 |
+
*.npy filter=lfs diff=lfs merge=lfs -text
|
| 15 |
+
*.npz filter=lfs diff=lfs merge=lfs -text
|
| 16 |
+
*.onnx filter=lfs diff=lfs merge=lfs -text
|
| 17 |
+
*.ot filter=lfs diff=lfs merge=lfs -text
|
| 18 |
+
*.parquet filter=lfs diff=lfs merge=lfs -text
|
| 19 |
+
*.pb filter=lfs diff=lfs merge=lfs -text
|
| 20 |
+
*.pickle filter=lfs diff=lfs merge=lfs -text
|
| 21 |
+
*.pkl filter=lfs diff=lfs merge=lfs -text
|
| 22 |
+
*.pt filter=lfs diff=lfs merge=lfs -text
|
| 23 |
+
*.pth filter=lfs diff=lfs merge=lfs -text
|
| 24 |
+
*.rar filter=lfs diff=lfs merge=lfs -text
|
| 25 |
+
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
| 26 |
+
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
| 27 |
+
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
| 28 |
+
*.tar filter=lfs diff=lfs merge=lfs -text
|
| 29 |
+
*.tflite filter=lfs diff=lfs merge=lfs -text
|
| 30 |
+
*.tgz filter=lfs diff=lfs merge=lfs -text
|
| 31 |
+
*.wasm filter=lfs diff=lfs merge=lfs -text
|
| 32 |
+
*.xz filter=lfs diff=lfs merge=lfs -text
|
| 33 |
+
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
+
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
+
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
README.md
ADDED
|
@@ -0,0 +1,175 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
base_model: perplexity-ai/pplx-embed-v2-late-9b
|
| 4 |
+
base_model_relation: finetune
|
| 5 |
+
library_name: transformers
|
| 6 |
+
pipeline_tag: feature-extraction
|
| 7 |
+
tags:
|
| 8 |
+
- feature-extraction
|
| 9 |
+
- sentence-similarity
|
| 10 |
+
- contextual-embeddings
|
| 11 |
+
- custom_code
|
| 12 |
+
language:
|
| 13 |
+
- multilingual
|
| 14 |
+
---
|
| 15 |
+
|
| 16 |
+
<p align="center">
|
| 17 |
+
<img src="assets/logo.svg" alt="Perplexity Logo" width="400">
|
| 18 |
+
</p>
|
| 19 |
+
|
| 20 |
+
<p align="center">pplx-embed-v2-context: Multimodal Contextual Embeddings</p>
|
| 21 |
+
|
| 22 |
+
`pplx-embed-v2-context-9b` is a state-of-the-art vision-language embedding model for retrieval over text, images, and visual documents.
|
| 23 |
+
It supports contextual and non-contextual embeddings: In contextual mode, the model embeds chunks of a document with the surrounding context in view. In non-contextual mode, text inputs are embedded in isolation.
|
| 24 |
+
Its 1024-dimensional MRL prefix shares an embedding space with [pplx-embed-v2-context-0.6b](https://huggingface.co/perplexity-ai/pplx-embed-v2-context-0.6b). Documents can be embedded with one model and queried with another.
|
| 25 |
+
|
| 26 |
+
This model also supports contextual chunking for visual inputs. Chunked images can be retrieved via standard dense ANN indices. Chunking images can substantially improve retrieval performance.
|
| 27 |
+
|
| 28 |
+
> [!IMPORTANT]
|
| 29 |
+
> When encoding chunks in contextual mode, the document chunks should be **non-overlapping**.
|
| 30 |
+
|
| 31 |
+
> [!IMPORTANT]
|
| 32 |
+
> Queries and documents are encoded with **different methods**: use `encode_queries` for queries and `encode` for document chunks. The model is trained with separate query and document prefixes, and encoding queries with `encode` silently degrades retrieval quality.
|
| 33 |
+
|
| 34 |
+
> [!IMPORTANT]
|
| 35 |
+
> Like `pplx-embed-context-v1`, the model natively produces *unnormalized* int8-quantized embeddings. Compare embeddings with *cosine similarity*, or pass `normalize_embeddings=True` and use the dot product.
|
| 36 |
+
|
| 37 |
+
## Text Embeddings
|
| 38 |
+
|
| 39 |
+
Requires `transformers>=5.4.0`, `torch`, `numpy`, `safetensors`, `tqdm`, Pillow, and torchvision. Install `flash-linear-attention` and `causal-conv1d` for the fast linear-attention path. The model uses custom code, so load it with `trust_remote_code=True`.
|
| 40 |
+
|
| 41 |
+
```python
|
| 42 |
+
from transformers import AutoModel
|
| 43 |
+
|
| 44 |
+
model = AutoModel.from_pretrained(
|
| 45 |
+
"perplexity-ai/pplx-embed-v2-context-9b",
|
| 46 |
+
trust_remote_code=True,
|
| 47 |
+
).to("cuda")
|
| 48 |
+
|
| 49 |
+
# Two documents, consisting of several chunks
|
| 50 |
+
documents = [
|
| 51 |
+
[
|
| 52 |
+
"Curiosity begins in childhood with endless questions about the world.",
|
| 53 |
+
"As we grow, curiosity drives us to explore new ideas.",
|
| 54 |
+
"Scientific breakthroughs often start with a curious question.",
|
| 55 |
+
],
|
| 56 |
+
[
|
| 57 |
+
"The curiosity rover explores Mars searching for ancient life.",
|
| 58 |
+
"Each discovery on Mars sparks new questions about the universe.",
|
| 59 |
+
],
|
| 60 |
+
]
|
| 61 |
+
|
| 62 |
+
# Contextual mode
|
| 63 |
+
# One (chunk_count, 2048) array per document:
|
| 64 |
+
# chunk_embeddings[0].shape == (3, 2048), chunk_embeddings[1].shape == (2, 2048)
|
| 65 |
+
chunk_embeddings = model.encode(documents, normalize_embeddings=True)
|
| 66 |
+
|
| 67 |
+
# Each query is a single-chunk row.
|
| 68 |
+
queries = [["What drives scientific breakthroughs?"]]
|
| 69 |
+
query_embeddings = model.encode_queries(queries, normalize_embeddings=True)
|
| 70 |
+
|
| 71 |
+
scores = chunk_embeddings[0] @ query_embeddings[0][0]
|
| 72 |
+
|
| 73 |
+
# Non-contextual mode: flatten chunks into a batch of singletons
|
| 74 |
+
chunk_embeddings = model.encode(
|
| 75 |
+
[[chunk] for doc in documents for chunk in doc],
|
| 76 |
+
normalize_embeddings=True,
|
| 77 |
+
)
|
| 78 |
+
```
|
| 79 |
+
|
| 80 |
+
`encode(documents, ...)` and `encode_queries(queries, ...)` accept:
|
| 81 |
+
|
| 82 |
+
| Argument | Default | Description |
|
| 83 |
+
|---|---|---|
|
| 84 |
+
| `batch_size` | `32` | Documents (or queries) per forward pass |
|
| 85 |
+
| `normalize_embeddings` | `False` | L2-normalize outputs |
|
| 86 |
+
| `convert_to_numpy` | `True` | Return NumPy arrays; `False` returns CPU tensors |
|
| 87 |
+
| `show_progress_bar` | `False` | Show a progress bar |
|
| 88 |
+
| `device` | `None` | Move the model to this device before encoding |
|
| 89 |
+
|
| 90 |
+
The text API does not truncate inputs. Split or truncate documents that exceed
|
| 91 |
+
the configured input length before encoding. To obtain one embedding per
|
| 92 |
+
document, pass each document as a single-chunk list.
|
| 93 |
+
|
| 94 |
+
### Matryoshka dimensions
|
| 95 |
+
|
| 96 |
+
The model was trained with Matryoshka losses at **1024** and **2048** dimensions. To use 1024-dimensional embeddings, take the first 1024 values of each unnormalized embedding and normalize afterwards:
|
| 97 |
+
|
| 98 |
+
```python
|
| 99 |
+
import numpy as np
|
| 100 |
+
|
| 101 |
+
emb = model.encode(documents) # unnormalized int8 values
|
| 102 |
+
emb_1024 = [e[:, :1024] / np.linalg.norm(e[:, :1024], axis=-1, keepdims=True) for e in emb]
|
| 103 |
+
```
|
| 104 |
+
|
| 105 |
+
Apply the same truncation and normalization to image embeddings and query
|
| 106 |
+
embeddings. Use 1024 dimensions when comparing against embeddings from the
|
| 107 |
+
0.6B model. Other truncation sizes were not trained.
|
| 108 |
+
|
| 109 |
+
## Image Embeddings
|
| 110 |
+
|
| 111 |
+
Use `encode_images()` to embed document pages or other images as contextual
|
| 112 |
+
chunks. Pass a list of PIL images; each image produces one embedding per
|
| 113 |
+
horizontal band. Use `encode_queries()` for the corresponding text queries.
|
| 114 |
+
|
| 115 |
+
```python
|
| 116 |
+
import torch
|
| 117 |
+
from PIL import Image
|
| 118 |
+
from transformers import AutoModel
|
| 119 |
+
|
| 120 |
+
model = AutoModel.from_pretrained(
|
| 121 |
+
"perplexity-ai/pplx-embed-v2-context-9b",
|
| 122 |
+
trust_remote_code=True,
|
| 123 |
+
dtype=torch.bfloat16,
|
| 124 |
+
attn_implementation="sdpa",
|
| 125 |
+
).to("cuda").eval()
|
| 126 |
+
|
| 127 |
+
with Image.open("page.png") as page:
|
| 128 |
+
image_embeddings = model.encode_images(
|
| 129 |
+
[page.convert("RGB")],
|
| 130 |
+
chunks=8,
|
| 131 |
+
min_pixels=16384,
|
| 132 |
+
max_pixels=1800000,
|
| 133 |
+
normalize_embeddings=True,
|
| 134 |
+
)
|
| 135 |
+
|
| 136 |
+
query_embeddings = model.encode_queries(
|
| 137 |
+
[["What is the company's annual revenue?"]],
|
| 138 |
+
normalize_embeddings=True,
|
| 139 |
+
)
|
| 140 |
+
page_score = (image_embeddings[0] @ query_embeddings[0][0]).max()
|
| 141 |
+
```
|
| 142 |
+
|
| 143 |
+
Each image is encoded in one contextual forward pass, with its merged visual
|
| 144 |
+
tokens divided into whole-row horizontal bands. Native three-axis MRoPE and image
|
| 145 |
+
wrappers are preserved. Separator tokens between bands use the midpoint of the
|
| 146 |
+
adjacent rows and the horizontal center of the image as their spatial position.
|
| 147 |
+
Only visual content is pooled; prefixes, wrappers, and separators are excluded.
|
| 148 |
+
|
| 149 |
+
`encode_images()` returns one `(actual_chunks, 2048)` array per page. The actual
|
| 150 |
+
chunk count is capped at the number of merged image rows.
|
| 151 |
+
Use cosine similarity per chunk and the maximum chunk score
|
| 152 |
+
for page retrieval, as shown above with normalized vectors.
|
| 153 |
+
|
| 154 |
+
| Image argument | Default | Description |
|
| 155 |
+
|---|---|---|
|
| 156 |
+
| `chunks` | `8` | Requested number of horizontal bands per page |
|
| 157 |
+
| `min_pixels` | `16384` | Processor minimum image pixel budget |
|
| 158 |
+
| `max_pixels` | `1800000` | Processor maximum image pixel budget |
|
| 159 |
+
| `batch_size` | `1` | Images per forward pass |
|
| 160 |
+
|
| 161 |
+
The image API also accepts `normalize_embeddings`, `convert_to_numpy`,
|
| 162 |
+
`show_progress_bar`, and `device` with the defaults listed above.
|
| 163 |
+
`prepare_images(images, chunks=8, min_pixels=16384, max_pixels=1800000)` exposes
|
| 164 |
+
the input tensors, chunk spans, masks, and position IDs without running the model.
|
| 165 |
+
|
| 166 |
+
## Citation
|
| 167 |
+
|
| 168 |
+
```bibtex
|
| 169 |
+
@online{perplexity2026ContextualEmbeddingBeyondTheGoldPassage,
|
| 170 |
+
author = {{Perplexity} and Sarah Eslami and William Barber and Markus Krimmel and Antoine Chaffin and Bo Wang and Louis Milliken and Jarrod Bryan and Claire Beamer and Jeremy Yang},
|
| 171 |
+
title = {{Contextual embedding beyond the gold passage}},
|
| 172 |
+
date = {2026-09-30},
|
| 173 |
+
url = {https://www.perplexity.ai/hub/blog/contextual-embedding-beyond-the-gold-passage}
|
| 174 |
+
}
|
| 175 |
+
```
|
assets/logo.svg
ADDED
|
config.json
ADDED
|
@@ -0,0 +1,120 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"PplxContextualModel"
|
| 4 |
+
],
|
| 5 |
+
"dtype": "float32",
|
| 6 |
+
"image_token_id": 248056,
|
| 7 |
+
"model_type": "pplx_contextual_qwen3_5",
|
| 8 |
+
"text_config": {
|
| 9 |
+
"attention_bias": false,
|
| 10 |
+
"attention_dropout": 0.0,
|
| 11 |
+
"attn_output_gate": true,
|
| 12 |
+
"bos_token_id": null,
|
| 13 |
+
"dtype": "float32",
|
| 14 |
+
"eos_token_id": 248044,
|
| 15 |
+
"full_attention_interval": 4,
|
| 16 |
+
"head_dim": 256,
|
| 17 |
+
"hidden_act": "silu",
|
| 18 |
+
"hidden_size": 4096,
|
| 19 |
+
"initializer_range": 0.02,
|
| 20 |
+
"intermediate_size": 12288,
|
| 21 |
+
"is_causal": false,
|
| 22 |
+
"layer_types": [
|
| 23 |
+
"linear_attention",
|
| 24 |
+
"linear_attention",
|
| 25 |
+
"linear_attention",
|
| 26 |
+
"full_attention",
|
| 27 |
+
"linear_attention",
|
| 28 |
+
"linear_attention",
|
| 29 |
+
"linear_attention",
|
| 30 |
+
"full_attention",
|
| 31 |
+
"linear_attention",
|
| 32 |
+
"linear_attention",
|
| 33 |
+
"linear_attention",
|
| 34 |
+
"full_attention",
|
| 35 |
+
"linear_attention",
|
| 36 |
+
"linear_attention",
|
| 37 |
+
"linear_attention",
|
| 38 |
+
"full_attention",
|
| 39 |
+
"linear_attention",
|
| 40 |
+
"linear_attention",
|
| 41 |
+
"linear_attention",
|
| 42 |
+
"full_attention",
|
| 43 |
+
"linear_attention",
|
| 44 |
+
"linear_attention",
|
| 45 |
+
"linear_attention",
|
| 46 |
+
"full_attention",
|
| 47 |
+
"linear_attention",
|
| 48 |
+
"linear_attention",
|
| 49 |
+
"linear_attention",
|
| 50 |
+
"full_attention",
|
| 51 |
+
"linear_attention",
|
| 52 |
+
"linear_attention",
|
| 53 |
+
"linear_attention",
|
| 54 |
+
"full_attention"
|
| 55 |
+
],
|
| 56 |
+
"linear_conv_kernel_dim": 4,
|
| 57 |
+
"linear_key_head_dim": 128,
|
| 58 |
+
"linear_num_key_heads": 16,
|
| 59 |
+
"linear_num_value_heads": 32,
|
| 60 |
+
"linear_value_head_dim": 128,
|
| 61 |
+
"mamba_ssm_dtype": "float32",
|
| 62 |
+
"max_position_embeddings": 262144,
|
| 63 |
+
"mlp_only_layers": [],
|
| 64 |
+
"model_type": "qwen3_5_text",
|
| 65 |
+
"mtp_num_hidden_layers": 1,
|
| 66 |
+
"mtp_use_dedicated_embeddings": false,
|
| 67 |
+
"num_attention_heads": 16,
|
| 68 |
+
"num_hidden_layers": 32,
|
| 69 |
+
"num_key_value_heads": 4,
|
| 70 |
+
"pad_token_id": null,
|
| 71 |
+
"partial_rotary_factor": 0.25,
|
| 72 |
+
"rms_norm_eps": 1e-06,
|
| 73 |
+
"rope_parameters": {
|
| 74 |
+
"mrope_interleaved": true,
|
| 75 |
+
"mrope_section": [
|
| 76 |
+
11,
|
| 77 |
+
11,
|
| 78 |
+
10
|
| 79 |
+
],
|
| 80 |
+
"partial_rotary_factor": 0.25,
|
| 81 |
+
"rope_theta": 10000000,
|
| 82 |
+
"rope_type": "default"
|
| 83 |
+
},
|
| 84 |
+
"tie_word_embeddings": false,
|
| 85 |
+
"use_cache": false,
|
| 86 |
+
"vocab_size": 248320
|
| 87 |
+
},
|
| 88 |
+
"tie_word_embeddings": false,
|
| 89 |
+
"transformers_version": "5.3.0",
|
| 90 |
+
"video_token_id": 248057,
|
| 91 |
+
"vision_config": {
|
| 92 |
+
"deepstack_visual_indexes": [],
|
| 93 |
+
"depth": 27,
|
| 94 |
+
"dtype": "float32",
|
| 95 |
+
"hidden_act": "gelu_pytorch_tanh",
|
| 96 |
+
"hidden_size": 1152,
|
| 97 |
+
"in_channels": 3,
|
| 98 |
+
"initializer_range": 0.02,
|
| 99 |
+
"intermediate_size": 4304,
|
| 100 |
+
"model_type": "qwen3_5",
|
| 101 |
+
"num_heads": 16,
|
| 102 |
+
"num_position_embeddings": 2304,
|
| 103 |
+
"out_hidden_size": 4096,
|
| 104 |
+
"patch_size": 16,
|
| 105 |
+
"spatial_merge_size": 2,
|
| 106 |
+
"temporal_patch_size": 2
|
| 107 |
+
},
|
| 108 |
+
"vision_end_token_id": 248054,
|
| 109 |
+
"vision_start_token_id": 248053,
|
| 110 |
+
"auto_map": {
|
| 111 |
+
"AutoConfig": "configuration_pplx_contextual.PplxContextualConfig",
|
| 112 |
+
"AutoModel": "modeling_pplx_contextual.PplxContextualModel"
|
| 113 |
+
},
|
| 114 |
+
"embedding_dim": 2048,
|
| 115 |
+
"query_prefix": "[Q] ",
|
| 116 |
+
"document_prefix": "[D] ",
|
| 117 |
+
"boundary_marker": "<|chunk_sep|>",
|
| 118 |
+
"query_length": 262144,
|
| 119 |
+
"document_length": 262144
|
| 120 |
+
}
|
configuration_pplx_contextual.py
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from typing import Any
|
| 2 |
+
|
| 3 |
+
from transformers import Qwen3_5Config
|
| 4 |
+
|
| 5 |
+
|
| 6 |
+
class PplxContextualConfig(Qwen3_5Config):
|
| 7 |
+
model_type = "pplx_contextual_qwen3_5"
|
| 8 |
+
|
| 9 |
+
def __init__(
|
| 10 |
+
self,
|
| 11 |
+
embedding_dim: int = 2048,
|
| 12 |
+
query_prefix: str = "[Q] ",
|
| 13 |
+
document_prefix: str = "[D] ",
|
| 14 |
+
boundary_marker: str = "<|chunk_sep|>",
|
| 15 |
+
query_length: int = 32768,
|
| 16 |
+
document_length: int = 32768,
|
| 17 |
+
**kwargs: Any,
|
| 18 |
+
) -> None:
|
| 19 |
+
super().__init__(**kwargs)
|
| 20 |
+
self.embedding_dim = embedding_dim
|
| 21 |
+
self.query_prefix = query_prefix
|
| 22 |
+
self.document_prefix = document_prefix
|
| 23 |
+
self.boundary_marker = boundary_marker
|
| 24 |
+
self.query_length = query_length
|
| 25 |
+
self.document_length = document_length
|
| 26 |
+
self.max_position_embeddings = document_length
|
| 27 |
+
self.text_config.use_cache = False
|
contextual_head.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:073d617eb35536b74598c8aaa3036ba7408c178b525ef27949e8074ffce6a51e
|
| 3 |
+
size 33554536
|
image_processing.py
ADDED
|
@@ -0,0 +1,146 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Contextual retrieval chunks from a whole image's merged visual tokens."""
|
| 2 |
+
|
| 3 |
+
from dataclasses import dataclass
|
| 4 |
+
|
| 5 |
+
import torch
|
| 6 |
+
from PIL import Image
|
| 7 |
+
from torch.nn.utils.rnn import pad_sequence
|
| 8 |
+
from transformers import BatchEncoding
|
| 9 |
+
|
| 10 |
+
@dataclass(frozen=True)
|
| 11 |
+
class ImageInputMapper:
|
| 12 |
+
_processor: object
|
| 13 |
+
tokenizer: object
|
| 14 |
+
document_prefix: str
|
| 15 |
+
boundary_marker: str
|
| 16 |
+
|
| 17 |
+
|
| 18 |
+
@dataclass(frozen=True)
|
| 19 |
+
class ImageEncodingConfig:
|
| 20 |
+
chunks: int = 8
|
| 21 |
+
min_pixels: int = 16384
|
| 22 |
+
max_pixels: int = 1800000
|
| 23 |
+
|
| 24 |
+
def __post_init__(self) -> None:
|
| 25 |
+
assert self.chunks > 0
|
| 26 |
+
assert 0 < self.min_pixels <= self.max_pixels
|
| 27 |
+
|
| 28 |
+
def chunk_indices(self, indices: torch.Tensor, grid: torch.Tensor, merge_size: int) -> list[torch.Tensor]:
|
| 29 |
+
frames, height, width = map(int, grid)
|
| 30 |
+
assert frames == 1 and height % merge_size == 0 and width % merge_size == 0
|
| 31 |
+
height, width = height // merge_size, width // merge_size
|
| 32 |
+
assert len(indices) == height * width, "Visual token count does not match merged image grid"
|
| 33 |
+
return [band.flatten() for band in indices.reshape(height, width).tensor_split(min(self.chunks, height), dim=0)]
|
| 34 |
+
|
| 35 |
+
def prepare_native_batch(self, mapper: ImageInputMapper, images: list[Image.Image]) -> BatchEncoding:
|
| 36 |
+
processor = mapper._processor
|
| 37 |
+
placeholder = processor.vision_start_token + processor.image_token + processor.vision_end_token
|
| 38 |
+
batch = BatchEncoding(processor(
|
| 39 |
+
text=[placeholder] * len(images),
|
| 40 |
+
images=[[image] for image in images],
|
| 41 |
+
return_tensors="pt",
|
| 42 |
+
padding=True,
|
| 43 |
+
min_pixels=self.min_pixels,
|
| 44 |
+
max_pixels=self.max_pixels,
|
| 45 |
+
))
|
| 46 |
+
prefix_id = mapper.tokenizer.convert_tokens_to_ids(mapper.document_prefix)
|
| 47 |
+
prefix = torch.full((len(images), 1), prefix_id, dtype=torch.long)
|
| 48 |
+
batch["input_ids"] = torch.cat((prefix, batch["input_ids"]), dim=1)
|
| 49 |
+
batch["attention_mask"] = torch.cat((torch.ones_like(prefix), batch["attention_mask"]), dim=1)
|
| 50 |
+
for key in ("mm_token_type_ids", "token_type_ids"):
|
| 51 |
+
if key in batch:
|
| 52 |
+
batch[key] = torch.cat((torch.zeros_like(prefix), batch[key]), dim=1)
|
| 53 |
+
assert "position_ids" not in batch
|
| 54 |
+
image_id = mapper.tokenizer.convert_tokens_to_ids(mapper._processor.image_token)
|
| 55 |
+
visual = batch["input_ids"] == image_id
|
| 56 |
+
assert torch.equal(batch["mm_token_type_ids"] == 1, visual)
|
| 57 |
+
spans = []
|
| 58 |
+
for row, grid in zip(visual, batch["image_grid_thw"], strict=True):
|
| 59 |
+
indices = row.nonzero().flatten()
|
| 60 |
+
assert torch.equal(indices, torch.arange(indices[0], indices[-1] + 1))
|
| 61 |
+
groups = self.chunk_indices(indices, grid, mapper._processor.image_processor.merge_size)
|
| 62 |
+
spans.append([(int(part[0]), int(part[-1]) + 1) for part in groups])
|
| 63 |
+
batch["chunk_token_spans"] = pad_sequence(
|
| 64 |
+
[torch.tensor(row, dtype=torch.long) for row in spans], batch_first=True
|
| 65 |
+
)
|
| 66 |
+
batch["chunk_mask"] = pad_sequence([torch.ones(len(row), dtype=torch.bool) for row in spans], batch_first=True)
|
| 67 |
+
return batch
|
| 68 |
+
|
| 69 |
+
def prepare(self, mapper: ImageInputMapper, images: list[Image.Image]) -> BatchEncoding:
|
| 70 |
+
batch = self.prepare_native_batch(mapper, images)
|
| 71 |
+
positions = torch.zeros(3, *batch["input_ids"].shape, dtype=torch.float32)
|
| 72 |
+
merge = mapper._processor.image_processor.merge_size
|
| 73 |
+
for row, grid in enumerate(batch["image_grid_thw"]):
|
| 74 |
+
active = batch["attention_mask"][row].bool()
|
| 75 |
+
visual = batch["mm_token_type_ids"][row, active] == 1
|
| 76 |
+
indices = visual.nonzero().flatten()
|
| 77 |
+
start, end = int(indices[0]), int(indices[-1]) + 1
|
| 78 |
+
height, width = int(grid[1]) // merge, int(grid[2]) // merge
|
| 79 |
+
native = torch.arange(int(active.sum())).float().expand(3, -1).clone()
|
| 80 |
+
native[:, start:end] = torch.stack(
|
| 81 |
+
(
|
| 82 |
+
torch.full((height * width,), start),
|
| 83 |
+
torch.arange(height).repeat_interleave(width) + start,
|
| 84 |
+
torch.arange(width).repeat(height) + start,
|
| 85 |
+
)
|
| 86 |
+
)
|
| 87 |
+
native[:, end:] = torch.arange(native.shape[1] - end) + start + max(height, width)
|
| 88 |
+
positions[:, row, active] = native
|
| 89 |
+
batch["position_ids"] = positions
|
| 90 |
+
return self.insert_native_separators(batch, mapper)
|
| 91 |
+
|
| 92 |
+
def boundary_positions(self, native: torch.Tensor, start: int, visual: torch.Tensor) -> torch.Tensor:
|
| 93 |
+
left, right = native[:, start - 1], native[:, start]
|
| 94 |
+
assert left[0] == right[0] and right[1] == left[1] + 1, "Separator must lie between adjacent image rows"
|
| 95 |
+
widths = native[2, visual]
|
| 96 |
+
return torch.stack((left[0], (left[1] + right[1]) / 2, (widths.min() + widths.max()) / 2))[:, None]
|
| 97 |
+
|
| 98 |
+
def insert_native_separators(self, batch: BatchEncoding, mapper: ImageInputMapper) -> BatchEncoding:
|
| 99 |
+
separator = mapper.tokenizer.encode(mapper.boundary_marker, add_special_tokens=False, split_special_tokens=True)
|
| 100 |
+
assert len(separator) == 1, "Image chunk boundaries require a single separator token"
|
| 101 |
+
assert batch["position_ids"].shape[0] == 3
|
| 102 |
+
token_keys = ["input_ids", "mm_token_type_ids"]
|
| 103 |
+
if "token_type_ids" in batch:
|
| 104 |
+
token_keys.append("token_type_ids")
|
| 105 |
+
rows = {key: [] for key in token_keys}
|
| 106 |
+
positions, spans = [], []
|
| 107 |
+
for row, mask in enumerate(batch["attention_mask"]):
|
| 108 |
+
active = mask.nonzero().flatten()
|
| 109 |
+
assert torch.equal(active, torch.arange(active[0], active[-1] + 1, device=active.device))
|
| 110 |
+
native = batch["position_ids"][:, row, active].float()
|
| 111 |
+
original = {key: batch[key][row, active] for key in token_keys}
|
| 112 |
+
parts = {key: [] for key in token_keys}
|
| 113 |
+
position_parts, row_spans = [], []
|
| 114 |
+
cursor, added = 0, 0
|
| 115 |
+
valid_spans = batch["chunk_token_spans"][row, batch["chunk_mask"][row]]
|
| 116 |
+
for index, (start, end) in enumerate(valid_spans.tolist()):
|
| 117 |
+
start, end = start - int(active[0]), end - int(active[0])
|
| 118 |
+
if index:
|
| 119 |
+
sep_positions = self.boundary_positions(native, start, original["mm_token_type_ids"] == 1)
|
| 120 |
+
count = sep_positions.shape[1]
|
| 121 |
+
for key in token_keys:
|
| 122 |
+
parts[key].append(original[key][cursor:start])
|
| 123 |
+
values = separator * (count // len(separator)) if key == "input_ids" else [0] * count
|
| 124 |
+
parts[key].append(original[key].new_tensor(values))
|
| 125 |
+
position_parts.extend((native[:, cursor:start], sep_positions))
|
| 126 |
+
added += count
|
| 127 |
+
cursor = start
|
| 128 |
+
row_spans.append((start + added, end + added))
|
| 129 |
+
for key in token_keys:
|
| 130 |
+
parts[key].append(original[key][cursor:])
|
| 131 |
+
rows[key].append(torch.cat(parts[key]))
|
| 132 |
+
position_parts.append(native[:, cursor:])
|
| 133 |
+
positions.append(torch.cat(position_parts, dim=1).T)
|
| 134 |
+
spans.append(row_spans)
|
| 135 |
+
for key in token_keys:
|
| 136 |
+
batch[key] = pad_sequence(
|
| 137 |
+
rows[key],
|
| 138 |
+
batch_first=True,
|
| 139 |
+
padding_value=mapper.tokenizer.pad_token_id if key == "input_ids" else 0,
|
| 140 |
+
)
|
| 141 |
+
batch["attention_mask"] = pad_sequence([torch.ones_like(row) for row in rows["input_ids"]], batch_first=True)
|
| 142 |
+
batch["position_ids"] = pad_sequence(positions, batch_first=True).permute(2, 0, 1)
|
| 143 |
+
batch["chunk_token_spans"] = pad_sequence(
|
| 144 |
+
[batch["chunk_token_spans"].new_tensor(row) for row in spans], batch_first=True
|
| 145 |
+
)
|
| 146 |
+
return batch
|
model-00001-of-00007.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e4dbe24849b8310e1ebad77ac2af380c5a707af35ba1dfd5d40a1883c1d5afb8
|
| 3 |
+
size 4943302192
|
model-00002-of-00007.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:786daf6875c2d1cf3aabd0ebbc8ab45383449a5efed8874a88c24af9c13b39bd
|
| 3 |
+
size 4803180760
|
model-00003-of-00007.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:60b5429f715daae9cf4f9643934a6ec53a6943f2f190881159e17eec304579f8
|
| 3 |
+
size 4970987816
|
model-00004-of-00007.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e30304189f45dad98b423aa228d3e8a0acca2bdde9b1a6a6e5ac98b78336118e
|
| 3 |
+
size 4938611312
|
model-00005-of-00007.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1b7012b55a946c2859a2f74514899e0cfb644b4c67bd6a7b1a43b4942d7b6057
|
| 3 |
+
size 4953012272
|
model-00006-of-00007.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7594d53d27c4b230a1a68ae06bd82c293bced745f2a0e667a5c345b6efd34f4b
|
| 3 |
+
size 4821156456
|
model-00007-of-00007.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:feffec312d10e3d3b25b39c2b62378ae28351a47a58fc3b50bd32e25c33ae208
|
| 3 |
+
size 4140617632
|
model.safetensors.index.json
ADDED
|
@@ -0,0 +1,767 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"metadata": {
|
| 3 |
+
"total_size": 33604334528
|
| 4 |
+
},
|
| 5 |
+
"weight_map": {
|
| 6 |
+
"language_model.embed_tokens.weight": "model-00001-of-00007.safetensors",
|
| 7 |
+
"language_model.layers.0.input_layernorm.weight": "model-00001-of-00007.safetensors",
|
| 8 |
+
"language_model.layers.0.linear_attn.A_log": "model-00001-of-00007.safetensors",
|
| 9 |
+
"language_model.layers.0.linear_attn.conv1d.weight": "model-00001-of-00007.safetensors",
|
| 10 |
+
"language_model.layers.0.linear_attn.dt_bias": "model-00001-of-00007.safetensors",
|
| 11 |
+
"language_model.layers.0.linear_attn.in_proj_a.weight": "model-00001-of-00007.safetensors",
|
| 12 |
+
"language_model.layers.0.linear_attn.in_proj_b.weight": "model-00001-of-00007.safetensors",
|
| 13 |
+
"language_model.layers.0.linear_attn.in_proj_qkv.weight": "model-00001-of-00007.safetensors",
|
| 14 |
+
"language_model.layers.0.linear_attn.in_proj_z.weight": "model-00001-of-00007.safetensors",
|
| 15 |
+
"language_model.layers.0.linear_attn.norm.weight": "model-00001-of-00007.safetensors",
|
| 16 |
+
"language_model.layers.0.linear_attn.out_proj.weight": "model-00001-of-00007.safetensors",
|
| 17 |
+
"language_model.layers.0.mlp.down_proj.weight": "model-00001-of-00007.safetensors",
|
| 18 |
+
"language_model.layers.0.mlp.gate_proj.weight": "model-00001-of-00007.safetensors",
|
| 19 |
+
"language_model.layers.0.mlp.up_proj.weight": "model-00001-of-00007.safetensors",
|
| 20 |
+
"language_model.layers.0.post_attention_layernorm.weight": "model-00001-of-00007.safetensors",
|
| 21 |
+
"language_model.layers.1.input_layernorm.weight": "model-00001-of-00007.safetensors",
|
| 22 |
+
"language_model.layers.1.linear_attn.A_log": "model-00001-of-00007.safetensors",
|
| 23 |
+
"language_model.layers.1.linear_attn.conv1d.weight": "model-00001-of-00007.safetensors",
|
| 24 |
+
"language_model.layers.1.linear_attn.dt_bias": "model-00001-of-00007.safetensors",
|
| 25 |
+
"language_model.layers.1.linear_attn.in_proj_a.weight": "model-00001-of-00007.safetensors",
|
| 26 |
+
"language_model.layers.1.linear_attn.in_proj_b.weight": "model-00001-of-00007.safetensors",
|
| 27 |
+
"language_model.layers.1.linear_attn.in_proj_qkv.weight": "model-00002-of-00007.safetensors",
|
| 28 |
+
"language_model.layers.1.linear_attn.in_proj_z.weight": "model-00002-of-00007.safetensors",
|
| 29 |
+
"language_model.layers.1.linear_attn.norm.weight": "model-00002-of-00007.safetensors",
|
| 30 |
+
"language_model.layers.1.linear_attn.out_proj.weight": "model-00002-of-00007.safetensors",
|
| 31 |
+
"language_model.layers.1.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
|
| 32 |
+
"language_model.layers.1.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
|
| 33 |
+
"language_model.layers.1.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
|
| 34 |
+
"language_model.layers.1.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
|
| 35 |
+
"language_model.layers.10.input_layernorm.weight": "model-00003-of-00007.safetensors",
|
| 36 |
+
"language_model.layers.10.linear_attn.A_log": "model-00003-of-00007.safetensors",
|
| 37 |
+
"language_model.layers.10.linear_attn.conv1d.weight": "model-00003-of-00007.safetensors",
|
| 38 |
+
"language_model.layers.10.linear_attn.dt_bias": "model-00003-of-00007.safetensors",
|
| 39 |
+
"language_model.layers.10.linear_attn.in_proj_a.weight": "model-00003-of-00007.safetensors",
|
| 40 |
+
"language_model.layers.10.linear_attn.in_proj_b.weight": "model-00003-of-00007.safetensors",
|
| 41 |
+
"language_model.layers.10.linear_attn.in_proj_qkv.weight": "model-00003-of-00007.safetensors",
|
| 42 |
+
"language_model.layers.10.linear_attn.in_proj_z.weight": "model-00003-of-00007.safetensors",
|
| 43 |
+
"language_model.layers.10.linear_attn.norm.weight": "model-00003-of-00007.safetensors",
|
| 44 |
+
"language_model.layers.10.linear_attn.out_proj.weight": "model-00003-of-00007.safetensors",
|
| 45 |
+
"language_model.layers.10.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
|
| 46 |
+
"language_model.layers.10.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
|
| 47 |
+
"language_model.layers.10.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
|
| 48 |
+
"language_model.layers.10.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
|
| 49 |
+
"language_model.layers.11.input_layernorm.weight": "model-00003-of-00007.safetensors",
|
| 50 |
+
"language_model.layers.11.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
|
| 51 |
+
"language_model.layers.11.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
|
| 52 |
+
"language_model.layers.11.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
|
| 53 |
+
"language_model.layers.11.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
|
| 54 |
+
"language_model.layers.11.self_attn.k_norm.weight": "model-00003-of-00007.safetensors",
|
| 55 |
+
"language_model.layers.11.self_attn.k_proj.weight": "model-00003-of-00007.safetensors",
|
| 56 |
+
"language_model.layers.11.self_attn.o_proj.weight": "model-00003-of-00007.safetensors",
|
| 57 |
+
"language_model.layers.11.self_attn.q_norm.weight": "model-00003-of-00007.safetensors",
|
| 58 |
+
"language_model.layers.11.self_attn.q_proj.weight": "model-00003-of-00007.safetensors",
|
| 59 |
+
"language_model.layers.11.self_attn.v_proj.weight": "model-00003-of-00007.safetensors",
|
| 60 |
+
"language_model.layers.12.input_layernorm.weight": "model-00003-of-00007.safetensors",
|
| 61 |
+
"language_model.layers.12.linear_attn.A_log": "model-00003-of-00007.safetensors",
|
| 62 |
+
"language_model.layers.12.linear_attn.conv1d.weight": "model-00003-of-00007.safetensors",
|
| 63 |
+
"language_model.layers.12.linear_attn.dt_bias": "model-00003-of-00007.safetensors",
|
| 64 |
+
"language_model.layers.12.linear_attn.in_proj_a.weight": "model-00003-of-00007.safetensors",
|
| 65 |
+
"language_model.layers.12.linear_attn.in_proj_b.weight": "model-00003-of-00007.safetensors",
|
| 66 |
+
"language_model.layers.12.linear_attn.in_proj_qkv.weight": "model-00003-of-00007.safetensors",
|
| 67 |
+
"language_model.layers.12.linear_attn.in_proj_z.weight": "model-00003-of-00007.safetensors",
|
| 68 |
+
"language_model.layers.12.linear_attn.norm.weight": "model-00003-of-00007.safetensors",
|
| 69 |
+
"language_model.layers.12.linear_attn.out_proj.weight": "model-00003-of-00007.safetensors",
|
| 70 |
+
"language_model.layers.12.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
|
| 71 |
+
"language_model.layers.12.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
|
| 72 |
+
"language_model.layers.12.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
|
| 73 |
+
"language_model.layers.12.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
|
| 74 |
+
"language_model.layers.13.input_layernorm.weight": "model-00004-of-00007.safetensors",
|
| 75 |
+
"language_model.layers.13.linear_attn.A_log": "model-00004-of-00007.safetensors",
|
| 76 |
+
"language_model.layers.13.linear_attn.conv1d.weight": "model-00004-of-00007.safetensors",
|
| 77 |
+
"language_model.layers.13.linear_attn.dt_bias": "model-00004-of-00007.safetensors",
|
| 78 |
+
"language_model.layers.13.linear_attn.in_proj_a.weight": "model-00004-of-00007.safetensors",
|
| 79 |
+
"language_model.layers.13.linear_attn.in_proj_b.weight": "model-00004-of-00007.safetensors",
|
| 80 |
+
"language_model.layers.13.linear_attn.in_proj_qkv.weight": "model-00004-of-00007.safetensors",
|
| 81 |
+
"language_model.layers.13.linear_attn.in_proj_z.weight": "model-00004-of-00007.safetensors",
|
| 82 |
+
"language_model.layers.13.linear_attn.norm.weight": "model-00004-of-00007.safetensors",
|
| 83 |
+
"language_model.layers.13.linear_attn.out_proj.weight": "model-00004-of-00007.safetensors",
|
| 84 |
+
"language_model.layers.13.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
|
| 85 |
+
"language_model.layers.13.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
|
| 86 |
+
"language_model.layers.13.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
|
| 87 |
+
"language_model.layers.13.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
|
| 88 |
+
"language_model.layers.14.input_layernorm.weight": "model-00004-of-00007.safetensors",
|
| 89 |
+
"language_model.layers.14.linear_attn.A_log": "model-00004-of-00007.safetensors",
|
| 90 |
+
"language_model.layers.14.linear_attn.conv1d.weight": "model-00004-of-00007.safetensors",
|
| 91 |
+
"language_model.layers.14.linear_attn.dt_bias": "model-00004-of-00007.safetensors",
|
| 92 |
+
"language_model.layers.14.linear_attn.in_proj_a.weight": "model-00004-of-00007.safetensors",
|
| 93 |
+
"language_model.layers.14.linear_attn.in_proj_b.weight": "model-00004-of-00007.safetensors",
|
| 94 |
+
"language_model.layers.14.linear_attn.in_proj_qkv.weight": "model-00004-of-00007.safetensors",
|
| 95 |
+
"language_model.layers.14.linear_attn.in_proj_z.weight": "model-00004-of-00007.safetensors",
|
| 96 |
+
"language_model.layers.14.linear_attn.norm.weight": "model-00004-of-00007.safetensors",
|
| 97 |
+
"language_model.layers.14.linear_attn.out_proj.weight": "model-00004-of-00007.safetensors",
|
| 98 |
+
"language_model.layers.14.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
|
| 99 |
+
"language_model.layers.14.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
|
| 100 |
+
"language_model.layers.14.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
|
| 101 |
+
"language_model.layers.14.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
|
| 102 |
+
"language_model.layers.15.input_layernorm.weight": "model-00004-of-00007.safetensors",
|
| 103 |
+
"language_model.layers.15.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
|
| 104 |
+
"language_model.layers.15.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
|
| 105 |
+
"language_model.layers.15.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
|
| 106 |
+
"language_model.layers.15.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
|
| 107 |
+
"language_model.layers.15.self_attn.k_norm.weight": "model-00004-of-00007.safetensors",
|
| 108 |
+
"language_model.layers.15.self_attn.k_proj.weight": "model-00004-of-00007.safetensors",
|
| 109 |
+
"language_model.layers.15.self_attn.o_proj.weight": "model-00004-of-00007.safetensors",
|
| 110 |
+
"language_model.layers.15.self_attn.q_norm.weight": "model-00004-of-00007.safetensors",
|
| 111 |
+
"language_model.layers.15.self_attn.q_proj.weight": "model-00004-of-00007.safetensors",
|
| 112 |
+
"language_model.layers.15.self_attn.v_proj.weight": "model-00004-of-00007.safetensors",
|
| 113 |
+
"language_model.layers.16.input_layernorm.weight": "model-00004-of-00007.safetensors",
|
| 114 |
+
"language_model.layers.16.linear_attn.A_log": "model-00004-of-00007.safetensors",
|
| 115 |
+
"language_model.layers.16.linear_attn.conv1d.weight": "model-00004-of-00007.safetensors",
|
| 116 |
+
"language_model.layers.16.linear_attn.dt_bias": "model-00004-of-00007.safetensors",
|
| 117 |
+
"language_model.layers.16.linear_attn.in_proj_a.weight": "model-00004-of-00007.safetensors",
|
| 118 |
+
"language_model.layers.16.linear_attn.in_proj_b.weight": "model-00004-of-00007.safetensors",
|
| 119 |
+
"language_model.layers.16.linear_attn.in_proj_qkv.weight": "model-00004-of-00007.safetensors",
|
| 120 |
+
"language_model.layers.16.linear_attn.in_proj_z.weight": "model-00004-of-00007.safetensors",
|
| 121 |
+
"language_model.layers.16.linear_attn.norm.weight": "model-00004-of-00007.safetensors",
|
| 122 |
+
"language_model.layers.16.linear_attn.out_proj.weight": "model-00004-of-00007.safetensors",
|
| 123 |
+
"language_model.layers.16.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
|
| 124 |
+
"language_model.layers.16.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
|
| 125 |
+
"language_model.layers.16.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
|
| 126 |
+
"language_model.layers.16.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
|
| 127 |
+
"language_model.layers.17.input_layernorm.weight": "model-00004-of-00007.safetensors",
|
| 128 |
+
"language_model.layers.17.linear_attn.A_log": "model-00004-of-00007.safetensors",
|
| 129 |
+
"language_model.layers.17.linear_attn.conv1d.weight": "model-00004-of-00007.safetensors",
|
| 130 |
+
"language_model.layers.17.linear_attn.dt_bias": "model-00004-of-00007.safetensors",
|
| 131 |
+
"language_model.layers.17.linear_attn.in_proj_a.weight": "model-00004-of-00007.safetensors",
|
| 132 |
+
"language_model.layers.17.linear_attn.in_proj_b.weight": "model-00004-of-00007.safetensors",
|
| 133 |
+
"language_model.layers.17.linear_attn.in_proj_qkv.weight": "model-00004-of-00007.safetensors",
|
| 134 |
+
"language_model.layers.17.linear_attn.in_proj_z.weight": "model-00004-of-00007.safetensors",
|
| 135 |
+
"language_model.layers.17.linear_attn.norm.weight": "model-00004-of-00007.safetensors",
|
| 136 |
+
"language_model.layers.17.linear_attn.out_proj.weight": "model-00004-of-00007.safetensors",
|
| 137 |
+
"language_model.layers.17.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
|
| 138 |
+
"language_model.layers.17.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
|
| 139 |
+
"language_model.layers.17.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
|
| 140 |
+
"language_model.layers.17.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
|
| 141 |
+
"language_model.layers.18.input_layernorm.weight": "model-00004-of-00007.safetensors",
|
| 142 |
+
"language_model.layers.18.linear_attn.A_log": "model-00004-of-00007.safetensors",
|
| 143 |
+
"language_model.layers.18.linear_attn.conv1d.weight": "model-00004-of-00007.safetensors",
|
| 144 |
+
"language_model.layers.18.linear_attn.dt_bias": "model-00004-of-00007.safetensors",
|
| 145 |
+
"language_model.layers.18.linear_attn.in_proj_a.weight": "model-00004-of-00007.safetensors",
|
| 146 |
+
"language_model.layers.18.linear_attn.in_proj_b.weight": "model-00004-of-00007.safetensors",
|
| 147 |
+
"language_model.layers.18.linear_attn.in_proj_qkv.weight": "model-00005-of-00007.safetensors",
|
| 148 |
+
"language_model.layers.18.linear_attn.in_proj_z.weight": "model-00005-of-00007.safetensors",
|
| 149 |
+
"language_model.layers.18.linear_attn.norm.weight": "model-00005-of-00007.safetensors",
|
| 150 |
+
"language_model.layers.18.linear_attn.out_proj.weight": "model-00005-of-00007.safetensors",
|
| 151 |
+
"language_model.layers.18.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
|
| 152 |
+
"language_model.layers.18.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
|
| 153 |
+
"language_model.layers.18.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
|
| 154 |
+
"language_model.layers.18.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
|
| 155 |
+
"language_model.layers.19.input_layernorm.weight": "model-00005-of-00007.safetensors",
|
| 156 |
+
"language_model.layers.19.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
|
| 157 |
+
"language_model.layers.19.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
|
| 158 |
+
"language_model.layers.19.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
|
| 159 |
+
"language_model.layers.19.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
|
| 160 |
+
"language_model.layers.19.self_attn.k_norm.weight": "model-00005-of-00007.safetensors",
|
| 161 |
+
"language_model.layers.19.self_attn.k_proj.weight": "model-00005-of-00007.safetensors",
|
| 162 |
+
"language_model.layers.19.self_attn.o_proj.weight": "model-00005-of-00007.safetensors",
|
| 163 |
+
"language_model.layers.19.self_attn.q_norm.weight": "model-00005-of-00007.safetensors",
|
| 164 |
+
"language_model.layers.19.self_attn.q_proj.weight": "model-00005-of-00007.safetensors",
|
| 165 |
+
"language_model.layers.19.self_attn.v_proj.weight": "model-00005-of-00007.safetensors",
|
| 166 |
+
"language_model.layers.2.input_layernorm.weight": "model-00002-of-00007.safetensors",
|
| 167 |
+
"language_model.layers.2.linear_attn.A_log": "model-00002-of-00007.safetensors",
|
| 168 |
+
"language_model.layers.2.linear_attn.conv1d.weight": "model-00002-of-00007.safetensors",
|
| 169 |
+
"language_model.layers.2.linear_attn.dt_bias": "model-00002-of-00007.safetensors",
|
| 170 |
+
"language_model.layers.2.linear_attn.in_proj_a.weight": "model-00002-of-00007.safetensors",
|
| 171 |
+
"language_model.layers.2.linear_attn.in_proj_b.weight": "model-00002-of-00007.safetensors",
|
| 172 |
+
"language_model.layers.2.linear_attn.in_proj_qkv.weight": "model-00002-of-00007.safetensors",
|
| 173 |
+
"language_model.layers.2.linear_attn.in_proj_z.weight": "model-00002-of-00007.safetensors",
|
| 174 |
+
"language_model.layers.2.linear_attn.norm.weight": "model-00002-of-00007.safetensors",
|
| 175 |
+
"language_model.layers.2.linear_attn.out_proj.weight": "model-00002-of-00007.safetensors",
|
| 176 |
+
"language_model.layers.2.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
|
| 177 |
+
"language_model.layers.2.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
|
| 178 |
+
"language_model.layers.2.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
|
| 179 |
+
"language_model.layers.2.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
|
| 180 |
+
"language_model.layers.20.input_layernorm.weight": "model-00005-of-00007.safetensors",
|
| 181 |
+
"language_model.layers.20.linear_attn.A_log": "model-00005-of-00007.safetensors",
|
| 182 |
+
"language_model.layers.20.linear_attn.conv1d.weight": "model-00005-of-00007.safetensors",
|
| 183 |
+
"language_model.layers.20.linear_attn.dt_bias": "model-00005-of-00007.safetensors",
|
| 184 |
+
"language_model.layers.20.linear_attn.in_proj_a.weight": "model-00005-of-00007.safetensors",
|
| 185 |
+
"language_model.layers.20.linear_attn.in_proj_b.weight": "model-00005-of-00007.safetensors",
|
| 186 |
+
"language_model.layers.20.linear_attn.in_proj_qkv.weight": "model-00005-of-00007.safetensors",
|
| 187 |
+
"language_model.layers.20.linear_attn.in_proj_z.weight": "model-00005-of-00007.safetensors",
|
| 188 |
+
"language_model.layers.20.linear_attn.norm.weight": "model-00005-of-00007.safetensors",
|
| 189 |
+
"language_model.layers.20.linear_attn.out_proj.weight": "model-00005-of-00007.safetensors",
|
| 190 |
+
"language_model.layers.20.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
|
| 191 |
+
"language_model.layers.20.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
|
| 192 |
+
"language_model.layers.20.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
|
| 193 |
+
"language_model.layers.20.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
|
| 194 |
+
"language_model.layers.21.input_layernorm.weight": "model-00005-of-00007.safetensors",
|
| 195 |
+
"language_model.layers.21.linear_attn.A_log": "model-00005-of-00007.safetensors",
|
| 196 |
+
"language_model.layers.21.linear_attn.conv1d.weight": "model-00005-of-00007.safetensors",
|
| 197 |
+
"language_model.layers.21.linear_attn.dt_bias": "model-00005-of-00007.safetensors",
|
| 198 |
+
"language_model.layers.21.linear_attn.in_proj_a.weight": "model-00005-of-00007.safetensors",
|
| 199 |
+
"language_model.layers.21.linear_attn.in_proj_b.weight": "model-00005-of-00007.safetensors",
|
| 200 |
+
"language_model.layers.21.linear_attn.in_proj_qkv.weight": "model-00005-of-00007.safetensors",
|
| 201 |
+
"language_model.layers.21.linear_attn.in_proj_z.weight": "model-00005-of-00007.safetensors",
|
| 202 |
+
"language_model.layers.21.linear_attn.norm.weight": "model-00005-of-00007.safetensors",
|
| 203 |
+
"language_model.layers.21.linear_attn.out_proj.weight": "model-00005-of-00007.safetensors",
|
| 204 |
+
"language_model.layers.21.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
|
| 205 |
+
"language_model.layers.21.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
|
| 206 |
+
"language_model.layers.21.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
|
| 207 |
+
"language_model.layers.21.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
|
| 208 |
+
"language_model.layers.22.input_layernorm.weight": "model-00005-of-00007.safetensors",
|
| 209 |
+
"language_model.layers.22.linear_attn.A_log": "model-00005-of-00007.safetensors",
|
| 210 |
+
"language_model.layers.22.linear_attn.conv1d.weight": "model-00005-of-00007.safetensors",
|
| 211 |
+
"language_model.layers.22.linear_attn.dt_bias": "model-00005-of-00007.safetensors",
|
| 212 |
+
"language_model.layers.22.linear_attn.in_proj_a.weight": "model-00005-of-00007.safetensors",
|
| 213 |
+
"language_model.layers.22.linear_attn.in_proj_b.weight": "model-00005-of-00007.safetensors",
|
| 214 |
+
"language_model.layers.22.linear_attn.in_proj_qkv.weight": "model-00005-of-00007.safetensors",
|
| 215 |
+
"language_model.layers.22.linear_attn.in_proj_z.weight": "model-00005-of-00007.safetensors",
|
| 216 |
+
"language_model.layers.22.linear_attn.norm.weight": "model-00005-of-00007.safetensors",
|
| 217 |
+
"language_model.layers.22.linear_attn.out_proj.weight": "model-00005-of-00007.safetensors",
|
| 218 |
+
"language_model.layers.22.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
|
| 219 |
+
"language_model.layers.22.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
|
| 220 |
+
"language_model.layers.22.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
|
| 221 |
+
"language_model.layers.22.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
|
| 222 |
+
"language_model.layers.23.input_layernorm.weight": "model-00005-of-00007.safetensors",
|
| 223 |
+
"language_model.layers.23.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
|
| 224 |
+
"language_model.layers.23.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
|
| 225 |
+
"language_model.layers.23.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
|
| 226 |
+
"language_model.layers.23.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
|
| 227 |
+
"language_model.layers.23.self_attn.k_norm.weight": "model-00005-of-00007.safetensors",
|
| 228 |
+
"language_model.layers.23.self_attn.k_proj.weight": "model-00005-of-00007.safetensors",
|
| 229 |
+
"language_model.layers.23.self_attn.o_proj.weight": "model-00006-of-00007.safetensors",
|
| 230 |
+
"language_model.layers.23.self_attn.q_norm.weight": "model-00006-of-00007.safetensors",
|
| 231 |
+
"language_model.layers.23.self_attn.q_proj.weight": "model-00006-of-00007.safetensors",
|
| 232 |
+
"language_model.layers.23.self_attn.v_proj.weight": "model-00006-of-00007.safetensors",
|
| 233 |
+
"language_model.layers.24.input_layernorm.weight": "model-00006-of-00007.safetensors",
|
| 234 |
+
"language_model.layers.24.linear_attn.A_log": "model-00006-of-00007.safetensors",
|
| 235 |
+
"language_model.layers.24.linear_attn.conv1d.weight": "model-00006-of-00007.safetensors",
|
| 236 |
+
"language_model.layers.24.linear_attn.dt_bias": "model-00006-of-00007.safetensors",
|
| 237 |
+
"language_model.layers.24.linear_attn.in_proj_a.weight": "model-00006-of-00007.safetensors",
|
| 238 |
+
"language_model.layers.24.linear_attn.in_proj_b.weight": "model-00006-of-00007.safetensors",
|
| 239 |
+
"language_model.layers.24.linear_attn.in_proj_qkv.weight": "model-00006-of-00007.safetensors",
|
| 240 |
+
"language_model.layers.24.linear_attn.in_proj_z.weight": "model-00006-of-00007.safetensors",
|
| 241 |
+
"language_model.layers.24.linear_attn.norm.weight": "model-00006-of-00007.safetensors",
|
| 242 |
+
"language_model.layers.24.linear_attn.out_proj.weight": "model-00006-of-00007.safetensors",
|
| 243 |
+
"language_model.layers.24.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
|
| 244 |
+
"language_model.layers.24.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
|
| 245 |
+
"language_model.layers.24.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
|
| 246 |
+
"language_model.layers.24.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
|
| 247 |
+
"language_model.layers.25.input_layernorm.weight": "model-00006-of-00007.safetensors",
|
| 248 |
+
"language_model.layers.25.linear_attn.A_log": "model-00006-of-00007.safetensors",
|
| 249 |
+
"language_model.layers.25.linear_attn.conv1d.weight": "model-00006-of-00007.safetensors",
|
| 250 |
+
"language_model.layers.25.linear_attn.dt_bias": "model-00006-of-00007.safetensors",
|
| 251 |
+
"language_model.layers.25.linear_attn.in_proj_a.weight": "model-00006-of-00007.safetensors",
|
| 252 |
+
"language_model.layers.25.linear_attn.in_proj_b.weight": "model-00006-of-00007.safetensors",
|
| 253 |
+
"language_model.layers.25.linear_attn.in_proj_qkv.weight": "model-00006-of-00007.safetensors",
|
| 254 |
+
"language_model.layers.25.linear_attn.in_proj_z.weight": "model-00006-of-00007.safetensors",
|
| 255 |
+
"language_model.layers.25.linear_attn.norm.weight": "model-00006-of-00007.safetensors",
|
| 256 |
+
"language_model.layers.25.linear_attn.out_proj.weight": "model-00006-of-00007.safetensors",
|
| 257 |
+
"language_model.layers.25.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
|
| 258 |
+
"language_model.layers.25.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
|
| 259 |
+
"language_model.layers.25.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
|
| 260 |
+
"language_model.layers.25.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
|
| 261 |
+
"language_model.layers.26.input_layernorm.weight": "model-00006-of-00007.safetensors",
|
| 262 |
+
"language_model.layers.26.linear_attn.A_log": "model-00006-of-00007.safetensors",
|
| 263 |
+
"language_model.layers.26.linear_attn.conv1d.weight": "model-00006-of-00007.safetensors",
|
| 264 |
+
"language_model.layers.26.linear_attn.dt_bias": "model-00006-of-00007.safetensors",
|
| 265 |
+
"language_model.layers.26.linear_attn.in_proj_a.weight": "model-00006-of-00007.safetensors",
|
| 266 |
+
"language_model.layers.26.linear_attn.in_proj_b.weight": "model-00006-of-00007.safetensors",
|
| 267 |
+
"language_model.layers.26.linear_attn.in_proj_qkv.weight": "model-00006-of-00007.safetensors",
|
| 268 |
+
"language_model.layers.26.linear_attn.in_proj_z.weight": "model-00006-of-00007.safetensors",
|
| 269 |
+
"language_model.layers.26.linear_attn.norm.weight": "model-00006-of-00007.safetensors",
|
| 270 |
+
"language_model.layers.26.linear_attn.out_proj.weight": "model-00006-of-00007.safetensors",
|
| 271 |
+
"language_model.layers.26.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
|
| 272 |
+
"language_model.layers.26.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
|
| 273 |
+
"language_model.layers.26.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
|
| 274 |
+
"language_model.layers.26.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
|
| 275 |
+
"language_model.layers.27.input_layernorm.weight": "model-00006-of-00007.safetensors",
|
| 276 |
+
"language_model.layers.27.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
|
| 277 |
+
"language_model.layers.27.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
|
| 278 |
+
"language_model.layers.27.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
|
| 279 |
+
"language_model.layers.27.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
|
| 280 |
+
"language_model.layers.27.self_attn.k_norm.weight": "model-00006-of-00007.safetensors",
|
| 281 |
+
"language_model.layers.27.self_attn.k_proj.weight": "model-00006-of-00007.safetensors",
|
| 282 |
+
"language_model.layers.27.self_attn.o_proj.weight": "model-00006-of-00007.safetensors",
|
| 283 |
+
"language_model.layers.27.self_attn.q_norm.weight": "model-00006-of-00007.safetensors",
|
| 284 |
+
"language_model.layers.27.self_attn.q_proj.weight": "model-00006-of-00007.safetensors",
|
| 285 |
+
"language_model.layers.27.self_attn.v_proj.weight": "model-00006-of-00007.safetensors",
|
| 286 |
+
"language_model.layers.28.input_layernorm.weight": "model-00006-of-00007.safetensors",
|
| 287 |
+
"language_model.layers.28.linear_attn.A_log": "model-00006-of-00007.safetensors",
|
| 288 |
+
"language_model.layers.28.linear_attn.conv1d.weight": "model-00006-of-00007.safetensors",
|
| 289 |
+
"language_model.layers.28.linear_attn.dt_bias": "model-00006-of-00007.safetensors",
|
| 290 |
+
"language_model.layers.28.linear_attn.in_proj_a.weight": "model-00006-of-00007.safetensors",
|
| 291 |
+
"language_model.layers.28.linear_attn.in_proj_b.weight": "model-00006-of-00007.safetensors",
|
| 292 |
+
"language_model.layers.28.linear_attn.in_proj_qkv.weight": "model-00006-of-00007.safetensors",
|
| 293 |
+
"language_model.layers.28.linear_attn.in_proj_z.weight": "model-00006-of-00007.safetensors",
|
| 294 |
+
"language_model.layers.28.linear_attn.norm.weight": "model-00006-of-00007.safetensors",
|
| 295 |
+
"language_model.layers.28.linear_attn.out_proj.weight": "model-00006-of-00007.safetensors",
|
| 296 |
+
"language_model.layers.28.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
|
| 297 |
+
"language_model.layers.28.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
|
| 298 |
+
"language_model.layers.28.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
|
| 299 |
+
"language_model.layers.28.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
|
| 300 |
+
"language_model.layers.29.input_layernorm.weight": "model-00006-of-00007.safetensors",
|
| 301 |
+
"language_model.layers.29.linear_attn.A_log": "model-00006-of-00007.safetensors",
|
| 302 |
+
"language_model.layers.29.linear_attn.conv1d.weight": "model-00006-of-00007.safetensors",
|
| 303 |
+
"language_model.layers.29.linear_attn.dt_bias": "model-00006-of-00007.safetensors",
|
| 304 |
+
"language_model.layers.29.linear_attn.in_proj_a.weight": "model-00006-of-00007.safetensors",
|
| 305 |
+
"language_model.layers.29.linear_attn.in_proj_b.weight": "model-00006-of-00007.safetensors",
|
| 306 |
+
"language_model.layers.29.linear_attn.in_proj_qkv.weight": "model-00006-of-00007.safetensors",
|
| 307 |
+
"language_model.layers.29.linear_attn.in_proj_z.weight": "model-00006-of-00007.safetensors",
|
| 308 |
+
"language_model.layers.29.linear_attn.norm.weight": "model-00006-of-00007.safetensors",
|
| 309 |
+
"language_model.layers.29.linear_attn.out_proj.weight": "model-00006-of-00007.safetensors",
|
| 310 |
+
"language_model.layers.29.mlp.down_proj.weight": "model-00007-of-00007.safetensors",
|
| 311 |
+
"language_model.layers.29.mlp.gate_proj.weight": "model-00007-of-00007.safetensors",
|
| 312 |
+
"language_model.layers.29.mlp.up_proj.weight": "model-00007-of-00007.safetensors",
|
| 313 |
+
"language_model.layers.29.post_attention_layernorm.weight": "model-00007-of-00007.safetensors",
|
| 314 |
+
"language_model.layers.3.input_layernorm.weight": "model-00002-of-00007.safetensors",
|
| 315 |
+
"language_model.layers.3.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
|
| 316 |
+
"language_model.layers.3.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
|
| 317 |
+
"language_model.layers.3.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
|
| 318 |
+
"language_model.layers.3.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
|
| 319 |
+
"language_model.layers.3.self_attn.k_norm.weight": "model-00002-of-00007.safetensors",
|
| 320 |
+
"language_model.layers.3.self_attn.k_proj.weight": "model-00002-of-00007.safetensors",
|
| 321 |
+
"language_model.layers.3.self_attn.o_proj.weight": "model-00002-of-00007.safetensors",
|
| 322 |
+
"language_model.layers.3.self_attn.q_norm.weight": "model-00002-of-00007.safetensors",
|
| 323 |
+
"language_model.layers.3.self_attn.q_proj.weight": "model-00002-of-00007.safetensors",
|
| 324 |
+
"language_model.layers.3.self_attn.v_proj.weight": "model-00002-of-00007.safetensors",
|
| 325 |
+
"language_model.layers.30.input_layernorm.weight": "model-00007-of-00007.safetensors",
|
| 326 |
+
"language_model.layers.30.linear_attn.A_log": "model-00007-of-00007.safetensors",
|
| 327 |
+
"language_model.layers.30.linear_attn.conv1d.weight": "model-00007-of-00007.safetensors",
|
| 328 |
+
"language_model.layers.30.linear_attn.dt_bias": "model-00007-of-00007.safetensors",
|
| 329 |
+
"language_model.layers.30.linear_attn.in_proj_a.weight": "model-00007-of-00007.safetensors",
|
| 330 |
+
"language_model.layers.30.linear_attn.in_proj_b.weight": "model-00007-of-00007.safetensors",
|
| 331 |
+
"language_model.layers.30.linear_attn.in_proj_qkv.weight": "model-00007-of-00007.safetensors",
|
| 332 |
+
"language_model.layers.30.linear_attn.in_proj_z.weight": "model-00007-of-00007.safetensors",
|
| 333 |
+
"language_model.layers.30.linear_attn.norm.weight": "model-00007-of-00007.safetensors",
|
| 334 |
+
"language_model.layers.30.linear_attn.out_proj.weight": "model-00007-of-00007.safetensors",
|
| 335 |
+
"language_model.layers.30.mlp.down_proj.weight": "model-00007-of-00007.safetensors",
|
| 336 |
+
"language_model.layers.30.mlp.gate_proj.weight": "model-00007-of-00007.safetensors",
|
| 337 |
+
"language_model.layers.30.mlp.up_proj.weight": "model-00007-of-00007.safetensors",
|
| 338 |
+
"language_model.layers.30.post_attention_layernorm.weight": "model-00007-of-00007.safetensors",
|
| 339 |
+
"language_model.layers.31.input_layernorm.weight": "model-00007-of-00007.safetensors",
|
| 340 |
+
"language_model.layers.31.mlp.down_proj.weight": "model-00007-of-00007.safetensors",
|
| 341 |
+
"language_model.layers.31.mlp.gate_proj.weight": "model-00007-of-00007.safetensors",
|
| 342 |
+
"language_model.layers.31.mlp.up_proj.weight": "model-00007-of-00007.safetensors",
|
| 343 |
+
"language_model.layers.31.post_attention_layernorm.weight": "model-00007-of-00007.safetensors",
|
| 344 |
+
"language_model.layers.31.self_attn.k_norm.weight": "model-00007-of-00007.safetensors",
|
| 345 |
+
"language_model.layers.31.self_attn.k_proj.weight": "model-00007-of-00007.safetensors",
|
| 346 |
+
"language_model.layers.31.self_attn.o_proj.weight": "model-00007-of-00007.safetensors",
|
| 347 |
+
"language_model.layers.31.self_attn.q_norm.weight": "model-00007-of-00007.safetensors",
|
| 348 |
+
"language_model.layers.31.self_attn.q_proj.weight": "model-00007-of-00007.safetensors",
|
| 349 |
+
"language_model.layers.31.self_attn.v_proj.weight": "model-00007-of-00007.safetensors",
|
| 350 |
+
"language_model.layers.4.input_layernorm.weight": "model-00002-of-00007.safetensors",
|
| 351 |
+
"language_model.layers.4.linear_attn.A_log": "model-00002-of-00007.safetensors",
|
| 352 |
+
"language_model.layers.4.linear_attn.conv1d.weight": "model-00002-of-00007.safetensors",
|
| 353 |
+
"language_model.layers.4.linear_attn.dt_bias": "model-00002-of-00007.safetensors",
|
| 354 |
+
"language_model.layers.4.linear_attn.in_proj_a.weight": "model-00002-of-00007.safetensors",
|
| 355 |
+
"language_model.layers.4.linear_attn.in_proj_b.weight": "model-00002-of-00007.safetensors",
|
| 356 |
+
"language_model.layers.4.linear_attn.in_proj_qkv.weight": "model-00002-of-00007.safetensors",
|
| 357 |
+
"language_model.layers.4.linear_attn.in_proj_z.weight": "model-00002-of-00007.safetensors",
|
| 358 |
+
"language_model.layers.4.linear_attn.norm.weight": "model-00002-of-00007.safetensors",
|
| 359 |
+
"language_model.layers.4.linear_attn.out_proj.weight": "model-00002-of-00007.safetensors",
|
| 360 |
+
"language_model.layers.4.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
|
| 361 |
+
"language_model.layers.4.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
|
| 362 |
+
"language_model.layers.4.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
|
| 363 |
+
"language_model.layers.4.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
|
| 364 |
+
"language_model.layers.5.input_layernorm.weight": "model-00002-of-00007.safetensors",
|
| 365 |
+
"language_model.layers.5.linear_attn.A_log": "model-00002-of-00007.safetensors",
|
| 366 |
+
"language_model.layers.5.linear_attn.conv1d.weight": "model-00002-of-00007.safetensors",
|
| 367 |
+
"language_model.layers.5.linear_attn.dt_bias": "model-00002-of-00007.safetensors",
|
| 368 |
+
"language_model.layers.5.linear_attn.in_proj_a.weight": "model-00002-of-00007.safetensors",
|
| 369 |
+
"language_model.layers.5.linear_attn.in_proj_b.weight": "model-00002-of-00007.safetensors",
|
| 370 |
+
"language_model.layers.5.linear_attn.in_proj_qkv.weight": "model-00002-of-00007.safetensors",
|
| 371 |
+
"language_model.layers.5.linear_attn.in_proj_z.weight": "model-00002-of-00007.safetensors",
|
| 372 |
+
"language_model.layers.5.linear_attn.norm.weight": "model-00002-of-00007.safetensors",
|
| 373 |
+
"language_model.layers.5.linear_attn.out_proj.weight": "model-00002-of-00007.safetensors",
|
| 374 |
+
"language_model.layers.5.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
|
| 375 |
+
"language_model.layers.5.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
|
| 376 |
+
"language_model.layers.5.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
|
| 377 |
+
"language_model.layers.5.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
|
| 378 |
+
"language_model.layers.6.input_layernorm.weight": "model-00002-of-00007.safetensors",
|
| 379 |
+
"language_model.layers.6.linear_attn.A_log": "model-00002-of-00007.safetensors",
|
| 380 |
+
"language_model.layers.6.linear_attn.conv1d.weight": "model-00002-of-00007.safetensors",
|
| 381 |
+
"language_model.layers.6.linear_attn.dt_bias": "model-00002-of-00007.safetensors",
|
| 382 |
+
"language_model.layers.6.linear_attn.in_proj_a.weight": "model-00002-of-00007.safetensors",
|
| 383 |
+
"language_model.layers.6.linear_attn.in_proj_b.weight": "model-00002-of-00007.safetensors",
|
| 384 |
+
"language_model.layers.6.linear_attn.in_proj_qkv.weight": "model-00002-of-00007.safetensors",
|
| 385 |
+
"language_model.layers.6.linear_attn.in_proj_z.weight": "model-00002-of-00007.safetensors",
|
| 386 |
+
"language_model.layers.6.linear_attn.norm.weight": "model-00002-of-00007.safetensors",
|
| 387 |
+
"language_model.layers.6.linear_attn.out_proj.weight": "model-00002-of-00007.safetensors",
|
| 388 |
+
"language_model.layers.6.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
|
| 389 |
+
"language_model.layers.6.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
|
| 390 |
+
"language_model.layers.6.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
|
| 391 |
+
"language_model.layers.6.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
|
| 392 |
+
"language_model.layers.7.input_layernorm.weight": "model-00003-of-00007.safetensors",
|
| 393 |
+
"language_model.layers.7.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
|
| 394 |
+
"language_model.layers.7.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
|
| 395 |
+
"language_model.layers.7.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
|
| 396 |
+
"language_model.layers.7.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
|
| 397 |
+
"language_model.layers.7.self_attn.k_norm.weight": "model-00003-of-00007.safetensors",
|
| 398 |
+
"language_model.layers.7.self_attn.k_proj.weight": "model-00003-of-00007.safetensors",
|
| 399 |
+
"language_model.layers.7.self_attn.o_proj.weight": "model-00003-of-00007.safetensors",
|
| 400 |
+
"language_model.layers.7.self_attn.q_norm.weight": "model-00003-of-00007.safetensors",
|
| 401 |
+
"language_model.layers.7.self_attn.q_proj.weight": "model-00003-of-00007.safetensors",
|
| 402 |
+
"language_model.layers.7.self_attn.v_proj.weight": "model-00003-of-00007.safetensors",
|
| 403 |
+
"language_model.layers.8.input_layernorm.weight": "model-00003-of-00007.safetensors",
|
| 404 |
+
"language_model.layers.8.linear_attn.A_log": "model-00003-of-00007.safetensors",
|
| 405 |
+
"language_model.layers.8.linear_attn.conv1d.weight": "model-00003-of-00007.safetensors",
|
| 406 |
+
"language_model.layers.8.linear_attn.dt_bias": "model-00003-of-00007.safetensors",
|
| 407 |
+
"language_model.layers.8.linear_attn.in_proj_a.weight": "model-00003-of-00007.safetensors",
|
| 408 |
+
"language_model.layers.8.linear_attn.in_proj_b.weight": "model-00003-of-00007.safetensors",
|
| 409 |
+
"language_model.layers.8.linear_attn.in_proj_qkv.weight": "model-00003-of-00007.safetensors",
|
| 410 |
+
"language_model.layers.8.linear_attn.in_proj_z.weight": "model-00003-of-00007.safetensors",
|
| 411 |
+
"language_model.layers.8.linear_attn.norm.weight": "model-00003-of-00007.safetensors",
|
| 412 |
+
"language_model.layers.8.linear_attn.out_proj.weight": "model-00003-of-00007.safetensors",
|
| 413 |
+
"language_model.layers.8.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
|
| 414 |
+
"language_model.layers.8.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
|
| 415 |
+
"language_model.layers.8.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
|
| 416 |
+
"language_model.layers.8.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
|
| 417 |
+
"language_model.layers.9.input_layernorm.weight": "model-00003-of-00007.safetensors",
|
| 418 |
+
"language_model.layers.9.linear_attn.A_log": "model-00003-of-00007.safetensors",
|
| 419 |
+
"language_model.layers.9.linear_attn.conv1d.weight": "model-00003-of-00007.safetensors",
|
| 420 |
+
"language_model.layers.9.linear_attn.dt_bias": "model-00003-of-00007.safetensors",
|
| 421 |
+
"language_model.layers.9.linear_attn.in_proj_a.weight": "model-00003-of-00007.safetensors",
|
| 422 |
+
"language_model.layers.9.linear_attn.in_proj_b.weight": "model-00003-of-00007.safetensors",
|
| 423 |
+
"language_model.layers.9.linear_attn.in_proj_qkv.weight": "model-00003-of-00007.safetensors",
|
| 424 |
+
"language_model.layers.9.linear_attn.in_proj_z.weight": "model-00003-of-00007.safetensors",
|
| 425 |
+
"language_model.layers.9.linear_attn.norm.weight": "model-00003-of-00007.safetensors",
|
| 426 |
+
"language_model.layers.9.linear_attn.out_proj.weight": "model-00003-of-00007.safetensors",
|
| 427 |
+
"language_model.layers.9.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
|
| 428 |
+
"language_model.layers.9.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
|
| 429 |
+
"language_model.layers.9.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
|
| 430 |
+
"language_model.layers.9.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
|
| 431 |
+
"language_model.norm.weight": "model-00007-of-00007.safetensors",
|
| 432 |
+
"visual.blocks.0.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 433 |
+
"visual.blocks.0.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 434 |
+
"visual.blocks.0.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 435 |
+
"visual.blocks.0.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 436 |
+
"visual.blocks.0.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 437 |
+
"visual.blocks.0.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 438 |
+
"visual.blocks.0.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 439 |
+
"visual.blocks.0.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 440 |
+
"visual.blocks.0.norm1.bias": "model-00007-of-00007.safetensors",
|
| 441 |
+
"visual.blocks.0.norm1.weight": "model-00007-of-00007.safetensors",
|
| 442 |
+
"visual.blocks.0.norm2.bias": "model-00007-of-00007.safetensors",
|
| 443 |
+
"visual.blocks.0.norm2.weight": "model-00007-of-00007.safetensors",
|
| 444 |
+
"visual.blocks.1.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 445 |
+
"visual.blocks.1.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 446 |
+
"visual.blocks.1.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 447 |
+
"visual.blocks.1.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 448 |
+
"visual.blocks.1.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 449 |
+
"visual.blocks.1.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 450 |
+
"visual.blocks.1.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 451 |
+
"visual.blocks.1.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 452 |
+
"visual.blocks.1.norm1.bias": "model-00007-of-00007.safetensors",
|
| 453 |
+
"visual.blocks.1.norm1.weight": "model-00007-of-00007.safetensors",
|
| 454 |
+
"visual.blocks.1.norm2.bias": "model-00007-of-00007.safetensors",
|
| 455 |
+
"visual.blocks.1.norm2.weight": "model-00007-of-00007.safetensors",
|
| 456 |
+
"visual.blocks.10.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 457 |
+
"visual.blocks.10.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 458 |
+
"visual.blocks.10.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 459 |
+
"visual.blocks.10.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 460 |
+
"visual.blocks.10.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 461 |
+
"visual.blocks.10.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 462 |
+
"visual.blocks.10.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 463 |
+
"visual.blocks.10.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 464 |
+
"visual.blocks.10.norm1.bias": "model-00007-of-00007.safetensors",
|
| 465 |
+
"visual.blocks.10.norm1.weight": "model-00007-of-00007.safetensors",
|
| 466 |
+
"visual.blocks.10.norm2.bias": "model-00007-of-00007.safetensors",
|
| 467 |
+
"visual.blocks.10.norm2.weight": "model-00007-of-00007.safetensors",
|
| 468 |
+
"visual.blocks.11.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 469 |
+
"visual.blocks.11.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 470 |
+
"visual.blocks.11.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 471 |
+
"visual.blocks.11.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 472 |
+
"visual.blocks.11.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 473 |
+
"visual.blocks.11.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 474 |
+
"visual.blocks.11.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 475 |
+
"visual.blocks.11.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 476 |
+
"visual.blocks.11.norm1.bias": "model-00007-of-00007.safetensors",
|
| 477 |
+
"visual.blocks.11.norm1.weight": "model-00007-of-00007.safetensors",
|
| 478 |
+
"visual.blocks.11.norm2.bias": "model-00007-of-00007.safetensors",
|
| 479 |
+
"visual.blocks.11.norm2.weight": "model-00007-of-00007.safetensors",
|
| 480 |
+
"visual.blocks.12.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 481 |
+
"visual.blocks.12.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 482 |
+
"visual.blocks.12.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 483 |
+
"visual.blocks.12.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 484 |
+
"visual.blocks.12.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 485 |
+
"visual.blocks.12.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 486 |
+
"visual.blocks.12.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 487 |
+
"visual.blocks.12.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 488 |
+
"visual.blocks.12.norm1.bias": "model-00007-of-00007.safetensors",
|
| 489 |
+
"visual.blocks.12.norm1.weight": "model-00007-of-00007.safetensors",
|
| 490 |
+
"visual.blocks.12.norm2.bias": "model-00007-of-00007.safetensors",
|
| 491 |
+
"visual.blocks.12.norm2.weight": "model-00007-of-00007.safetensors",
|
| 492 |
+
"visual.blocks.13.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 493 |
+
"visual.blocks.13.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 494 |
+
"visual.blocks.13.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 495 |
+
"visual.blocks.13.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 496 |
+
"visual.blocks.13.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 497 |
+
"visual.blocks.13.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 498 |
+
"visual.blocks.13.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 499 |
+
"visual.blocks.13.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 500 |
+
"visual.blocks.13.norm1.bias": "model-00007-of-00007.safetensors",
|
| 501 |
+
"visual.blocks.13.norm1.weight": "model-00007-of-00007.safetensors",
|
| 502 |
+
"visual.blocks.13.norm2.bias": "model-00007-of-00007.safetensors",
|
| 503 |
+
"visual.blocks.13.norm2.weight": "model-00007-of-00007.safetensors",
|
| 504 |
+
"visual.blocks.14.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 505 |
+
"visual.blocks.14.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 506 |
+
"visual.blocks.14.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 507 |
+
"visual.blocks.14.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 508 |
+
"visual.blocks.14.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 509 |
+
"visual.blocks.14.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 510 |
+
"visual.blocks.14.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 511 |
+
"visual.blocks.14.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 512 |
+
"visual.blocks.14.norm1.bias": "model-00007-of-00007.safetensors",
|
| 513 |
+
"visual.blocks.14.norm1.weight": "model-00007-of-00007.safetensors",
|
| 514 |
+
"visual.blocks.14.norm2.bias": "model-00007-of-00007.safetensors",
|
| 515 |
+
"visual.blocks.14.norm2.weight": "model-00007-of-00007.safetensors",
|
| 516 |
+
"visual.blocks.15.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 517 |
+
"visual.blocks.15.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 518 |
+
"visual.blocks.15.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 519 |
+
"visual.blocks.15.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 520 |
+
"visual.blocks.15.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 521 |
+
"visual.blocks.15.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 522 |
+
"visual.blocks.15.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 523 |
+
"visual.blocks.15.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 524 |
+
"visual.blocks.15.norm1.bias": "model-00007-of-00007.safetensors",
|
| 525 |
+
"visual.blocks.15.norm1.weight": "model-00007-of-00007.safetensors",
|
| 526 |
+
"visual.blocks.15.norm2.bias": "model-00007-of-00007.safetensors",
|
| 527 |
+
"visual.blocks.15.norm2.weight": "model-00007-of-00007.safetensors",
|
| 528 |
+
"visual.blocks.16.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 529 |
+
"visual.blocks.16.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 530 |
+
"visual.blocks.16.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 531 |
+
"visual.blocks.16.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 532 |
+
"visual.blocks.16.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 533 |
+
"visual.blocks.16.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 534 |
+
"visual.blocks.16.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 535 |
+
"visual.blocks.16.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 536 |
+
"visual.blocks.16.norm1.bias": "model-00007-of-00007.safetensors",
|
| 537 |
+
"visual.blocks.16.norm1.weight": "model-00007-of-00007.safetensors",
|
| 538 |
+
"visual.blocks.16.norm2.bias": "model-00007-of-00007.safetensors",
|
| 539 |
+
"visual.blocks.16.norm2.weight": "model-00007-of-00007.safetensors",
|
| 540 |
+
"visual.blocks.17.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 541 |
+
"visual.blocks.17.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 542 |
+
"visual.blocks.17.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 543 |
+
"visual.blocks.17.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 544 |
+
"visual.blocks.17.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 545 |
+
"visual.blocks.17.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 546 |
+
"visual.blocks.17.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 547 |
+
"visual.blocks.17.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 548 |
+
"visual.blocks.17.norm1.bias": "model-00007-of-00007.safetensors",
|
| 549 |
+
"visual.blocks.17.norm1.weight": "model-00007-of-00007.safetensors",
|
| 550 |
+
"visual.blocks.17.norm2.bias": "model-00007-of-00007.safetensors",
|
| 551 |
+
"visual.blocks.17.norm2.weight": "model-00007-of-00007.safetensors",
|
| 552 |
+
"visual.blocks.18.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 553 |
+
"visual.blocks.18.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 554 |
+
"visual.blocks.18.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 555 |
+
"visual.blocks.18.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 556 |
+
"visual.blocks.18.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 557 |
+
"visual.blocks.18.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 558 |
+
"visual.blocks.18.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 559 |
+
"visual.blocks.18.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 560 |
+
"visual.blocks.18.norm1.bias": "model-00007-of-00007.safetensors",
|
| 561 |
+
"visual.blocks.18.norm1.weight": "model-00007-of-00007.safetensors",
|
| 562 |
+
"visual.blocks.18.norm2.bias": "model-00007-of-00007.safetensors",
|
| 563 |
+
"visual.blocks.18.norm2.weight": "model-00007-of-00007.safetensors",
|
| 564 |
+
"visual.blocks.19.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 565 |
+
"visual.blocks.19.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 566 |
+
"visual.blocks.19.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 567 |
+
"visual.blocks.19.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 568 |
+
"visual.blocks.19.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 569 |
+
"visual.blocks.19.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 570 |
+
"visual.blocks.19.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 571 |
+
"visual.blocks.19.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 572 |
+
"visual.blocks.19.norm1.bias": "model-00007-of-00007.safetensors",
|
| 573 |
+
"visual.blocks.19.norm1.weight": "model-00007-of-00007.safetensors",
|
| 574 |
+
"visual.blocks.19.norm2.bias": "model-00007-of-00007.safetensors",
|
| 575 |
+
"visual.blocks.19.norm2.weight": "model-00007-of-00007.safetensors",
|
| 576 |
+
"visual.blocks.2.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 577 |
+
"visual.blocks.2.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 578 |
+
"visual.blocks.2.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 579 |
+
"visual.blocks.2.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 580 |
+
"visual.blocks.2.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 581 |
+
"visual.blocks.2.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 582 |
+
"visual.blocks.2.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 583 |
+
"visual.blocks.2.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 584 |
+
"visual.blocks.2.norm1.bias": "model-00007-of-00007.safetensors",
|
| 585 |
+
"visual.blocks.2.norm1.weight": "model-00007-of-00007.safetensors",
|
| 586 |
+
"visual.blocks.2.norm2.bias": "model-00007-of-00007.safetensors",
|
| 587 |
+
"visual.blocks.2.norm2.weight": "model-00007-of-00007.safetensors",
|
| 588 |
+
"visual.blocks.20.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 589 |
+
"visual.blocks.20.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 590 |
+
"visual.blocks.20.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 591 |
+
"visual.blocks.20.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 592 |
+
"visual.blocks.20.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 593 |
+
"visual.blocks.20.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 594 |
+
"visual.blocks.20.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 595 |
+
"visual.blocks.20.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 596 |
+
"visual.blocks.20.norm1.bias": "model-00007-of-00007.safetensors",
|
| 597 |
+
"visual.blocks.20.norm1.weight": "model-00007-of-00007.safetensors",
|
| 598 |
+
"visual.blocks.20.norm2.bias": "model-00007-of-00007.safetensors",
|
| 599 |
+
"visual.blocks.20.norm2.weight": "model-00007-of-00007.safetensors",
|
| 600 |
+
"visual.blocks.21.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 601 |
+
"visual.blocks.21.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 602 |
+
"visual.blocks.21.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 603 |
+
"visual.blocks.21.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 604 |
+
"visual.blocks.21.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 605 |
+
"visual.blocks.21.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 606 |
+
"visual.blocks.21.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 607 |
+
"visual.blocks.21.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 608 |
+
"visual.blocks.21.norm1.bias": "model-00007-of-00007.safetensors",
|
| 609 |
+
"visual.blocks.21.norm1.weight": "model-00007-of-00007.safetensors",
|
| 610 |
+
"visual.blocks.21.norm2.bias": "model-00007-of-00007.safetensors",
|
| 611 |
+
"visual.blocks.21.norm2.weight": "model-00007-of-00007.safetensors",
|
| 612 |
+
"visual.blocks.22.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 613 |
+
"visual.blocks.22.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 614 |
+
"visual.blocks.22.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 615 |
+
"visual.blocks.22.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 616 |
+
"visual.blocks.22.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 617 |
+
"visual.blocks.22.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 618 |
+
"visual.blocks.22.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 619 |
+
"visual.blocks.22.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 620 |
+
"visual.blocks.22.norm1.bias": "model-00007-of-00007.safetensors",
|
| 621 |
+
"visual.blocks.22.norm1.weight": "model-00007-of-00007.safetensors",
|
| 622 |
+
"visual.blocks.22.norm2.bias": "model-00007-of-00007.safetensors",
|
| 623 |
+
"visual.blocks.22.norm2.weight": "model-00007-of-00007.safetensors",
|
| 624 |
+
"visual.blocks.23.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 625 |
+
"visual.blocks.23.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 626 |
+
"visual.blocks.23.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 627 |
+
"visual.blocks.23.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 628 |
+
"visual.blocks.23.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 629 |
+
"visual.blocks.23.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 630 |
+
"visual.blocks.23.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 631 |
+
"visual.blocks.23.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 632 |
+
"visual.blocks.23.norm1.bias": "model-00007-of-00007.safetensors",
|
| 633 |
+
"visual.blocks.23.norm1.weight": "model-00007-of-00007.safetensors",
|
| 634 |
+
"visual.blocks.23.norm2.bias": "model-00007-of-00007.safetensors",
|
| 635 |
+
"visual.blocks.23.norm2.weight": "model-00007-of-00007.safetensors",
|
| 636 |
+
"visual.blocks.24.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 637 |
+
"visual.blocks.24.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 638 |
+
"visual.blocks.24.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 639 |
+
"visual.blocks.24.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 640 |
+
"visual.blocks.24.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 641 |
+
"visual.blocks.24.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 642 |
+
"visual.blocks.24.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 643 |
+
"visual.blocks.24.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 644 |
+
"visual.blocks.24.norm1.bias": "model-00007-of-00007.safetensors",
|
| 645 |
+
"visual.blocks.24.norm1.weight": "model-00007-of-00007.safetensors",
|
| 646 |
+
"visual.blocks.24.norm2.bias": "model-00007-of-00007.safetensors",
|
| 647 |
+
"visual.blocks.24.norm2.weight": "model-00007-of-00007.safetensors",
|
| 648 |
+
"visual.blocks.25.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 649 |
+
"visual.blocks.25.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 650 |
+
"visual.blocks.25.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 651 |
+
"visual.blocks.25.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 652 |
+
"visual.blocks.25.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 653 |
+
"visual.blocks.25.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 654 |
+
"visual.blocks.25.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 655 |
+
"visual.blocks.25.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 656 |
+
"visual.blocks.25.norm1.bias": "model-00007-of-00007.safetensors",
|
| 657 |
+
"visual.blocks.25.norm1.weight": "model-00007-of-00007.safetensors",
|
| 658 |
+
"visual.blocks.25.norm2.bias": "model-00007-of-00007.safetensors",
|
| 659 |
+
"visual.blocks.25.norm2.weight": "model-00007-of-00007.safetensors",
|
| 660 |
+
"visual.blocks.26.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 661 |
+
"visual.blocks.26.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 662 |
+
"visual.blocks.26.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 663 |
+
"visual.blocks.26.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 664 |
+
"visual.blocks.26.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 665 |
+
"visual.blocks.26.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 666 |
+
"visual.blocks.26.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 667 |
+
"visual.blocks.26.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 668 |
+
"visual.blocks.26.norm1.bias": "model-00007-of-00007.safetensors",
|
| 669 |
+
"visual.blocks.26.norm1.weight": "model-00007-of-00007.safetensors",
|
| 670 |
+
"visual.blocks.26.norm2.bias": "model-00007-of-00007.safetensors",
|
| 671 |
+
"visual.blocks.26.norm2.weight": "model-00007-of-00007.safetensors",
|
| 672 |
+
"visual.blocks.3.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 673 |
+
"visual.blocks.3.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 674 |
+
"visual.blocks.3.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 675 |
+
"visual.blocks.3.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 676 |
+
"visual.blocks.3.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 677 |
+
"visual.blocks.3.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 678 |
+
"visual.blocks.3.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 679 |
+
"visual.blocks.3.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 680 |
+
"visual.blocks.3.norm1.bias": "model-00007-of-00007.safetensors",
|
| 681 |
+
"visual.blocks.3.norm1.weight": "model-00007-of-00007.safetensors",
|
| 682 |
+
"visual.blocks.3.norm2.bias": "model-00007-of-00007.safetensors",
|
| 683 |
+
"visual.blocks.3.norm2.weight": "model-00007-of-00007.safetensors",
|
| 684 |
+
"visual.blocks.4.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 685 |
+
"visual.blocks.4.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 686 |
+
"visual.blocks.4.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 687 |
+
"visual.blocks.4.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 688 |
+
"visual.blocks.4.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 689 |
+
"visual.blocks.4.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 690 |
+
"visual.blocks.4.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 691 |
+
"visual.blocks.4.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 692 |
+
"visual.blocks.4.norm1.bias": "model-00007-of-00007.safetensors",
|
| 693 |
+
"visual.blocks.4.norm1.weight": "model-00007-of-00007.safetensors",
|
| 694 |
+
"visual.blocks.4.norm2.bias": "model-00007-of-00007.safetensors",
|
| 695 |
+
"visual.blocks.4.norm2.weight": "model-00007-of-00007.safetensors",
|
| 696 |
+
"visual.blocks.5.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 697 |
+
"visual.blocks.5.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 698 |
+
"visual.blocks.5.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 699 |
+
"visual.blocks.5.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 700 |
+
"visual.blocks.5.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 701 |
+
"visual.blocks.5.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 702 |
+
"visual.blocks.5.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 703 |
+
"visual.blocks.5.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 704 |
+
"visual.blocks.5.norm1.bias": "model-00007-of-00007.safetensors",
|
| 705 |
+
"visual.blocks.5.norm1.weight": "model-00007-of-00007.safetensors",
|
| 706 |
+
"visual.blocks.5.norm2.bias": "model-00007-of-00007.safetensors",
|
| 707 |
+
"visual.blocks.5.norm2.weight": "model-00007-of-00007.safetensors",
|
| 708 |
+
"visual.blocks.6.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 709 |
+
"visual.blocks.6.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 710 |
+
"visual.blocks.6.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 711 |
+
"visual.blocks.6.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 712 |
+
"visual.blocks.6.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 713 |
+
"visual.blocks.6.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 714 |
+
"visual.blocks.6.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 715 |
+
"visual.blocks.6.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 716 |
+
"visual.blocks.6.norm1.bias": "model-00007-of-00007.safetensors",
|
| 717 |
+
"visual.blocks.6.norm1.weight": "model-00007-of-00007.safetensors",
|
| 718 |
+
"visual.blocks.6.norm2.bias": "model-00007-of-00007.safetensors",
|
| 719 |
+
"visual.blocks.6.norm2.weight": "model-00007-of-00007.safetensors",
|
| 720 |
+
"visual.blocks.7.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 721 |
+
"visual.blocks.7.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 722 |
+
"visual.blocks.7.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 723 |
+
"visual.blocks.7.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 724 |
+
"visual.blocks.7.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 725 |
+
"visual.blocks.7.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 726 |
+
"visual.blocks.7.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 727 |
+
"visual.blocks.7.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 728 |
+
"visual.blocks.7.norm1.bias": "model-00007-of-00007.safetensors",
|
| 729 |
+
"visual.blocks.7.norm1.weight": "model-00007-of-00007.safetensors",
|
| 730 |
+
"visual.blocks.7.norm2.bias": "model-00007-of-00007.safetensors",
|
| 731 |
+
"visual.blocks.7.norm2.weight": "model-00007-of-00007.safetensors",
|
| 732 |
+
"visual.blocks.8.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 733 |
+
"visual.blocks.8.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 734 |
+
"visual.blocks.8.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 735 |
+
"visual.blocks.8.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 736 |
+
"visual.blocks.8.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 737 |
+
"visual.blocks.8.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 738 |
+
"visual.blocks.8.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 739 |
+
"visual.blocks.8.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 740 |
+
"visual.blocks.8.norm1.bias": "model-00007-of-00007.safetensors",
|
| 741 |
+
"visual.blocks.8.norm1.weight": "model-00007-of-00007.safetensors",
|
| 742 |
+
"visual.blocks.8.norm2.bias": "model-00007-of-00007.safetensors",
|
| 743 |
+
"visual.blocks.8.norm2.weight": "model-00007-of-00007.safetensors",
|
| 744 |
+
"visual.blocks.9.attn.proj.bias": "model-00007-of-00007.safetensors",
|
| 745 |
+
"visual.blocks.9.attn.proj.weight": "model-00007-of-00007.safetensors",
|
| 746 |
+
"visual.blocks.9.attn.qkv.bias": "model-00007-of-00007.safetensors",
|
| 747 |
+
"visual.blocks.9.attn.qkv.weight": "model-00007-of-00007.safetensors",
|
| 748 |
+
"visual.blocks.9.mlp.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 749 |
+
"visual.blocks.9.mlp.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 750 |
+
"visual.blocks.9.mlp.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 751 |
+
"visual.blocks.9.mlp.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 752 |
+
"visual.blocks.9.norm1.bias": "model-00007-of-00007.safetensors",
|
| 753 |
+
"visual.blocks.9.norm1.weight": "model-00007-of-00007.safetensors",
|
| 754 |
+
"visual.blocks.9.norm2.bias": "model-00007-of-00007.safetensors",
|
| 755 |
+
"visual.blocks.9.norm2.weight": "model-00007-of-00007.safetensors",
|
| 756 |
+
"visual.merger.linear_fc1.bias": "model-00007-of-00007.safetensors",
|
| 757 |
+
"visual.merger.linear_fc1.weight": "model-00007-of-00007.safetensors",
|
| 758 |
+
"visual.merger.linear_fc2.bias": "model-00007-of-00007.safetensors",
|
| 759 |
+
"visual.merger.linear_fc2.weight": "model-00007-of-00007.safetensors",
|
| 760 |
+
"visual.merger.norm.bias": "model-00007-of-00007.safetensors",
|
| 761 |
+
"visual.merger.norm.weight": "model-00007-of-00007.safetensors",
|
| 762 |
+
"visual.patch_embed.proj.bias": "model-00007-of-00007.safetensors",
|
| 763 |
+
"visual.patch_embed.proj.weight": "model-00007-of-00007.safetensors",
|
| 764 |
+
"visual.pos_embed.weight": "model-00007-of-00007.safetensors",
|
| 765 |
+
"contextual_projection.weight": "contextual_head.safetensors"
|
| 766 |
+
}
|
| 767 |
+
}
|
modeling_pplx_contextual.py
ADDED
|
@@ -0,0 +1,307 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from __future__ import annotations
|
| 2 |
+
|
| 3 |
+
from pathlib import Path
|
| 4 |
+
from typing import Any, Literal
|
| 5 |
+
|
| 6 |
+
import numpy as np
|
| 7 |
+
import torch
|
| 8 |
+
import torch.nn.functional as F
|
| 9 |
+
from torch import nn
|
| 10 |
+
from tqdm.auto import tqdm
|
| 11 |
+
from PIL import Image
|
| 12 |
+
from transformers import AutoTokenizer, PreTrainedTokenizerBase, Qwen3_5Model, Qwen3VLProcessor
|
| 13 |
+
|
| 14 |
+
from .configuration_pplx_contextual import PplxContextualConfig
|
| 15 |
+
from .image_processing import ImageEncodingConfig, ImageInputMapper
|
| 16 |
+
|
| 17 |
+
|
| 18 |
+
class PplxContextualModel(Qwen3_5Model):
|
| 19 |
+
"""Encode a document's supplied chunks together, then pool each chunk."""
|
| 20 |
+
|
| 21 |
+
config_class = PplxContextualConfig
|
| 22 |
+
config: PplxContextualConfig
|
| 23 |
+
_keep_in_fp32_modules = ["contextual_projection"]
|
| 24 |
+
|
| 25 |
+
def __init__(self, config: PplxContextualConfig) -> None:
|
| 26 |
+
super().__init__(config)
|
| 27 |
+
self.contextual_projection = nn.Linear(
|
| 28 |
+
config.text_config.hidden_size, config.embedding_dim, bias=False
|
| 29 |
+
)
|
| 30 |
+
self._tokenizer: PreTrainedTokenizerBase | None = None
|
| 31 |
+
self._image_processor: Qwen3VLProcessor | None = None
|
| 32 |
+
self.post_init()
|
| 33 |
+
|
| 34 |
+
@property
|
| 35 |
+
def tokenizer(self) -> PreTrainedTokenizerBase:
|
| 36 |
+
if self._tokenizer is None:
|
| 37 |
+
self._tokenizer = AutoTokenizer.from_pretrained(
|
| 38 |
+
self.config._name_or_path,
|
| 39 |
+
config=self.config,
|
| 40 |
+
revision=self.config._commit_hash,
|
| 41 |
+
padding_side="right",
|
| 42 |
+
)
|
| 43 |
+
return self._tokenizer
|
| 44 |
+
|
| 45 |
+
@tokenizer.setter
|
| 46 |
+
def tokenizer(self, value: PreTrainedTokenizerBase) -> None:
|
| 47 |
+
self._tokenizer = value
|
| 48 |
+
|
| 49 |
+
def prepare_inputs(
|
| 50 |
+
self, documents: list[list[str]], task: str = "document"
|
| 51 |
+
) -> dict[str, torch.Tensor]:
|
| 52 |
+
tokenizer = self.tokenizer
|
| 53 |
+
if task == "query":
|
| 54 |
+
if any(len(row) != 1 for row in documents):
|
| 55 |
+
raise ValueError("Each query must contain exactly one string")
|
| 56 |
+
batch = tokenizer(
|
| 57 |
+
[row[0] for row in documents],
|
| 58 |
+
padding=True,
|
| 59 |
+
truncation=False,
|
| 60 |
+
split_special_tokens=True,
|
| 61 |
+
return_tensors="pt",
|
| 62 |
+
return_token_type_ids=False,
|
| 63 |
+
)
|
| 64 |
+
prefix_id = tokenizer.convert_tokens_to_ids(
|
| 65 |
+
self.config.query_prefix
|
| 66 |
+
)
|
| 67 |
+
prefix = torch.full(
|
| 68 |
+
(len(documents), 1), prefix_id, dtype=torch.long
|
| 69 |
+
)
|
| 70 |
+
batch["input_ids"] = torch.cat((prefix, batch["input_ids"]), dim=1)
|
| 71 |
+
batch["attention_mask"] = torch.cat(
|
| 72 |
+
(torch.ones_like(prefix), batch["attention_mask"]), dim=1
|
| 73 |
+
)
|
| 74 |
+
batch["chunk_mask"] = torch.ones(
|
| 75 |
+
(len(documents), 1), dtype=torch.bool
|
| 76 |
+
)
|
| 77 |
+
elif task == "document":
|
| 78 |
+
texts, char_spans = [], []
|
| 79 |
+
for chunks in documents:
|
| 80 |
+
parts = [self.config.document_prefix]
|
| 81 |
+
cursor = len(self.config.document_prefix)
|
| 82 |
+
spans = []
|
| 83 |
+
for index, chunk in enumerate(chunks):
|
| 84 |
+
if index:
|
| 85 |
+
parts.append(self.config.boundary_marker)
|
| 86 |
+
cursor += len(self.config.boundary_marker)
|
| 87 |
+
spans.append((cursor, cursor + len(chunk)))
|
| 88 |
+
parts.append(chunk)
|
| 89 |
+
cursor += len(chunk)
|
| 90 |
+
texts.append("".join(parts))
|
| 91 |
+
char_spans.append(spans)
|
| 92 |
+
batch = tokenizer(
|
| 93 |
+
texts,
|
| 94 |
+
padding=True,
|
| 95 |
+
truncation=False,
|
| 96 |
+
split_special_tokens=True,
|
| 97 |
+
return_tensors="pt",
|
| 98 |
+
return_token_type_ids=False,
|
| 99 |
+
return_offsets_mapping=True,
|
| 100 |
+
)
|
| 101 |
+
offsets = batch.pop("offset_mapping")
|
| 102 |
+
count = max(1, *map(len, documents))
|
| 103 |
+
token_spans = torch.zeros(
|
| 104 |
+
(len(documents), count, 2), dtype=torch.long
|
| 105 |
+
)
|
| 106 |
+
chunk_mask = torch.zeros((len(documents), count), dtype=torch.bool)
|
| 107 |
+
for row, spans in enumerate(char_spans):
|
| 108 |
+
starts, ends = offsets[row].unbind(dim=-1)
|
| 109 |
+
valid = batch["attention_mask"][row].bool() & (ends > starts)
|
| 110 |
+
for column, (start, end) in enumerate(spans):
|
| 111 |
+
indices = (
|
| 112 |
+
(valid & (starts < end) & (ends > start))
|
| 113 |
+
.nonzero()
|
| 114 |
+
.flatten()
|
| 115 |
+
)
|
| 116 |
+
if indices.numel():
|
| 117 |
+
token_spans[row, column, 0] = indices[0]
|
| 118 |
+
token_spans[row, column, 1] = indices[-1] + 1
|
| 119 |
+
elif documents[row][column]:
|
| 120 |
+
raise ValueError(
|
| 121 |
+
f"Chunk {column} in document {row} produced no tokens"
|
| 122 |
+
)
|
| 123 |
+
chunk_mask[row, column] = True
|
| 124 |
+
batch["chunk_token_spans"] = token_spans
|
| 125 |
+
batch["chunk_mask"] = chunk_mask
|
| 126 |
+
else:
|
| 127 |
+
raise ValueError(f"Unsupported task: {task}")
|
| 128 |
+
limit = (
|
| 129 |
+
self.config.query_length
|
| 130 |
+
if task == "query"
|
| 131 |
+
else self.config.document_length
|
| 132 |
+
)
|
| 133 |
+
if batch["input_ids"].shape[1] > limit:
|
| 134 |
+
raise ValueError(
|
| 135 |
+
f"Input exceeds the {limit}-token limit; split the document into context windows"
|
| 136 |
+
)
|
| 137 |
+
return dict(batch)
|
| 138 |
+
|
| 139 |
+
def _pool(
|
| 140 |
+
self, features: dict[str, torch.Tensor], task: str
|
| 141 |
+
) -> torch.Tensor:
|
| 142 |
+
visual = {key: features[key] for key in (
|
| 143 |
+
"pixel_values", "image_grid_thw", "mm_token_type_ids", "position_ids"
|
| 144 |
+
) if key in features}
|
| 145 |
+
hidden = self(
|
| 146 |
+
input_ids=features["input_ids"],
|
| 147 |
+
attention_mask=features["attention_mask"],
|
| 148 |
+
use_cache=False,
|
| 149 |
+
return_dict=True,
|
| 150 |
+
**visual,
|
| 151 |
+
).last_hidden_state
|
| 152 |
+
with torch.autocast(device_type=hidden.device.type, enabled=False):
|
| 153 |
+
hidden = hidden.float()
|
| 154 |
+
if "pixel_values" in features:
|
| 155 |
+
hidden = F.linear(hidden, self.contextual_projection.weight.float())
|
| 156 |
+
if task == "query":
|
| 157 |
+
mask = features["attention_mask"].unsqueeze(-1).float()
|
| 158 |
+
pooled = (
|
| 159 |
+
(hidden * mask).sum(dim=1) / mask.sum(dim=1).clamp_min(1.0)
|
| 160 |
+
).unsqueeze(1)
|
| 161 |
+
else:
|
| 162 |
+
prefix = torch.cat(
|
| 163 |
+
(
|
| 164 |
+
hidden.new_zeros(hidden.shape[0], 1, hidden.shape[-1]),
|
| 165 |
+
hidden.cumsum(dim=1),
|
| 166 |
+
),
|
| 167 |
+
dim=1,
|
| 168 |
+
)
|
| 169 |
+
starts, ends = features["chunk_token_spans"].unbind(dim=-1)
|
| 170 |
+
rows = torch.arange(
|
| 171 |
+
hidden.shape[0], device=hidden.device
|
| 172 |
+
).unsqueeze(1)
|
| 173 |
+
pooled = (prefix[rows, ends] - prefix[rows, starts]) / (
|
| 174 |
+
ends - starts
|
| 175 |
+
).clamp_min(1).unsqueeze(-1)
|
| 176 |
+
if "pixel_values" in features:
|
| 177 |
+
return pooled
|
| 178 |
+
return F.linear(pooled, self.contextual_projection.weight.float())
|
| 179 |
+
|
| 180 |
+
def prepare_images(
|
| 181 |
+
self,
|
| 182 |
+
images: list[Image.Image],
|
| 183 |
+
*,
|
| 184 |
+
chunks: int = 8,
|
| 185 |
+
min_pixels: int = 16384,
|
| 186 |
+
max_pixels: int = 1800000,
|
| 187 |
+
) -> dict[str, torch.Tensor]:
|
| 188 |
+
"""Wrap whole pages and split their visual tokens into native-MRoPE row bands."""
|
| 189 |
+
if not images or any(not isinstance(image, Image.Image) for image in images):
|
| 190 |
+
raise ValueError("Expected a nonempty list of PIL images")
|
| 191 |
+
if self._image_processor is None:
|
| 192 |
+
self._image_processor = Qwen3VLProcessor.from_pretrained(
|
| 193 |
+
self.config._name_or_path, revision=self.config._commit_hash, trust_remote_code=True,
|
| 194 |
+
)
|
| 195 |
+
self._image_processor.tokenizer = self.tokenizer
|
| 196 |
+
mapper = ImageInputMapper(
|
| 197 |
+
self._image_processor, self.tokenizer,
|
| 198 |
+
self.config.document_prefix, self.config.boundary_marker,
|
| 199 |
+
)
|
| 200 |
+
features = ImageEncodingConfig(chunks, min_pixels, max_pixels).prepare(mapper, images)
|
| 201 |
+
if features["input_ids"].shape[1] > self.config.document_length:
|
| 202 |
+
raise ValueError("Image input exceeds document_length; reduce max_pixels")
|
| 203 |
+
return dict(features)
|
| 204 |
+
|
| 205 |
+
@torch.inference_mode()
|
| 206 |
+
def encode_images(
|
| 207 |
+
self,
|
| 208 |
+
images: list[Image.Image],
|
| 209 |
+
batch_size: int = 1,
|
| 210 |
+
show_progress_bar: bool = False,
|
| 211 |
+
device: str | torch.device | None = None,
|
| 212 |
+
normalize_embeddings: bool = False,
|
| 213 |
+
convert_to_numpy: bool = True,
|
| 214 |
+
quantization: Literal["int8"] = "int8",
|
| 215 |
+
*,
|
| 216 |
+
chunks: int = 8,
|
| 217 |
+
min_pixels: int = 16384,
|
| 218 |
+
max_pixels: int = 1800000,
|
| 219 |
+
) -> list[np.ndarray] | list[torch.Tensor]:
|
| 220 |
+
"""Return one quantized vector per nonempty row band of each page."""
|
| 221 |
+
if batch_size < 1 or quantization != "int8":
|
| 222 |
+
raise ValueError("Expected positive batch_size and int8 quantization")
|
| 223 |
+
self.eval()
|
| 224 |
+
if device is not None:
|
| 225 |
+
self.to(device)
|
| 226 |
+
rows = []
|
| 227 |
+
for start in tqdm(range(0, len(images), batch_size), disable=not show_progress_bar):
|
| 228 |
+
features = {
|
| 229 |
+
key: value.to(self.device)
|
| 230 |
+
for key, value in self.prepare_images(
|
| 231 |
+
images[start:start + batch_size], chunks=chunks,
|
| 232 |
+
min_pixels=min_pixels, max_pixels=max_pixels,
|
| 233 |
+
).items()
|
| 234 |
+
}
|
| 235 |
+
projected = self._pool(features, "document")
|
| 236 |
+
vectors = torch.round(torch.tanh(projected) * 127).clamp(-128, 127)
|
| 237 |
+
if normalize_embeddings:
|
| 238 |
+
vectors = F.normalize(vectors, p=2, dim=-1)
|
| 239 |
+
rows.extend(vector[mask].cpu() for vector, mask in zip(vectors, features["chunk_mask"], strict=True))
|
| 240 |
+
return [row.numpy() for row in rows] if convert_to_numpy else rows
|
| 241 |
+
|
| 242 |
+
@torch.inference_mode()
|
| 243 |
+
def encode(
|
| 244 |
+
self,
|
| 245 |
+
documents: list[list[str]],
|
| 246 |
+
batch_size: int = 32,
|
| 247 |
+
show_progress_bar: bool = False,
|
| 248 |
+
device: str | torch.device | None = None,
|
| 249 |
+
normalize_embeddings: bool = False,
|
| 250 |
+
convert_to_numpy: bool = True,
|
| 251 |
+
quantization: Literal["int8"] = "int8",
|
| 252 |
+
*,
|
| 253 |
+
task: Literal["query", "document"] = "document",
|
| 254 |
+
) -> list[np.ndarray] | list[torch.Tensor]:
|
| 255 |
+
"""Return one (chunk_count, embedding_dim) array per document.
|
| 256 |
+
|
| 257 |
+
Int8 quantization returns float32 integer values, matching the V1 API.
|
| 258 |
+
No automatic chunk splitting or input truncation is performed.
|
| 259 |
+
"""
|
| 260 |
+
if not isinstance(documents, list) or any(
|
| 261 |
+
not isinstance(row, list)
|
| 262 |
+
or any(not isinstance(text, str) for text in row)
|
| 263 |
+
for row in documents
|
| 264 |
+
):
|
| 265 |
+
raise TypeError(
|
| 266 |
+
"Expected a list of documents, each a list of text chunks"
|
| 267 |
+
)
|
| 268 |
+
if batch_size < 1:
|
| 269 |
+
raise ValueError("batch_size must be positive")
|
| 270 |
+
if task not in ("query", "document"):
|
| 271 |
+
raise ValueError(f"Unsupported task: {task}")
|
| 272 |
+
if quantization != "int8":
|
| 273 |
+
raise ValueError(f"Unsupported quantization: {quantization}")
|
| 274 |
+
self.eval()
|
| 275 |
+
if device is not None:
|
| 276 |
+
self.to(device)
|
| 277 |
+
rows = []
|
| 278 |
+
for start in tqdm(
|
| 279 |
+
range(0, len(documents), batch_size), disable=not show_progress_bar
|
| 280 |
+
):
|
| 281 |
+
batch = documents[start : start + batch_size]
|
| 282 |
+
features = {
|
| 283 |
+
key: value.to(self.device)
|
| 284 |
+
for key, value in self.prepare_inputs(batch, task).items()
|
| 285 |
+
}
|
| 286 |
+
projected = self._pool(features, task)
|
| 287 |
+
vectors = torch.round(torch.tanh(projected) * 127).clamp(-128, 127)
|
| 288 |
+
if normalize_embeddings:
|
| 289 |
+
vectors = F.normalize(vectors, p=2, dim=-1)
|
| 290 |
+
rows.extend(
|
| 291 |
+
vectors[index, : len(chunks)].cpu()
|
| 292 |
+
for index, chunks in enumerate(batch)
|
| 293 |
+
)
|
| 294 |
+
return [row.numpy() for row in rows] if convert_to_numpy else rows
|
| 295 |
+
|
| 296 |
+
def encode_queries(
|
| 297 |
+
self, queries: list[list[str]], **kwargs: Any
|
| 298 |
+
) -> list[np.ndarray] | list[torch.Tensor]:
|
| 299 |
+
"""Encode singleton query rows with the checkpoint's query prefix."""
|
| 300 |
+
return self.encode(queries, task="query", **kwargs)
|
| 301 |
+
|
| 302 |
+
def save_pretrained(
|
| 303 |
+
self, save_directory: str | Path, **kwargs: Any
|
| 304 |
+
) -> None:
|
| 305 |
+
tokenizer = self.tokenizer
|
| 306 |
+
super().save_pretrained(save_directory, **kwargs)
|
| 307 |
+
tokenizer.save_pretrained(save_directory)
|
processor_config.json
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"image_processor": {
|
| 3 |
+
"data_format": "channels_first",
|
| 4 |
+
"do_convert_rgb": true,
|
| 5 |
+
"do_normalize": true,
|
| 6 |
+
"do_rescale": true,
|
| 7 |
+
"do_resize": true,
|
| 8 |
+
"image_mean": [
|
| 9 |
+
0.5,
|
| 10 |
+
0.5,
|
| 11 |
+
0.5
|
| 12 |
+
],
|
| 13 |
+
"image_processor_type": "Qwen2VLImageProcessorFast",
|
| 14 |
+
"image_std": [
|
| 15 |
+
0.5,
|
| 16 |
+
0.5,
|
| 17 |
+
0.5
|
| 18 |
+
],
|
| 19 |
+
"merge_size": 2,
|
| 20 |
+
"patch_size": 16,
|
| 21 |
+
"resample": 3,
|
| 22 |
+
"rescale_factor": 0.00392156862745098,
|
| 23 |
+
"size": {
|
| 24 |
+
"longest_edge": 16777216,
|
| 25 |
+
"shortest_edge": 65536
|
| 26 |
+
},
|
| 27 |
+
"temporal_patch_size": 2
|
| 28 |
+
},
|
| 29 |
+
"processor_class": "Qwen3VLProcessor",
|
| 30 |
+
"video_processor": {
|
| 31 |
+
"data_format": "channels_first",
|
| 32 |
+
"default_to_square": true,
|
| 33 |
+
"do_convert_rgb": true,
|
| 34 |
+
"do_normalize": true,
|
| 35 |
+
"do_rescale": true,
|
| 36 |
+
"do_resize": true,
|
| 37 |
+
"do_sample_frames": true,
|
| 38 |
+
"fps": 2,
|
| 39 |
+
"image_mean": [
|
| 40 |
+
0.5,
|
| 41 |
+
0.5,
|
| 42 |
+
0.5
|
| 43 |
+
],
|
| 44 |
+
"image_std": [
|
| 45 |
+
0.5,
|
| 46 |
+
0.5,
|
| 47 |
+
0.5
|
| 48 |
+
],
|
| 49 |
+
"max_frames": 768,
|
| 50 |
+
"merge_size": 2,
|
| 51 |
+
"min_frames": 4,
|
| 52 |
+
"patch_size": 16,
|
| 53 |
+
"resample": 3,
|
| 54 |
+
"rescale_factor": 0.00392156862745098,
|
| 55 |
+
"return_metadata": false,
|
| 56 |
+
"size": {
|
| 57 |
+
"longest_edge": 25165824,
|
| 58 |
+
"shortest_edge": 4096
|
| 59 |
+
},
|
| 60 |
+
"temporal_patch_size": 2,
|
| 61 |
+
"video_processor_type": "Qwen3VLVideoProcessor"
|
| 62 |
+
}
|
| 63 |
+
}
|
tokenizer.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d76b678db762df7761041239ca8950fb190135c94bb66ce8aeff5a0ff2933cf5
|
| 3 |
+
size 19989896
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"audio_bos_token": "<|audio_start|>",
|
| 4 |
+
"audio_eos_token": "<|audio_end|>",
|
| 5 |
+
"audio_token": "<|audio_pad|>",
|
| 6 |
+
"backend": "tokenizers",
|
| 7 |
+
"bos_token": null,
|
| 8 |
+
"clean_up_tokenization_spaces": false,
|
| 9 |
+
"eos_token": "<|im_end|>",
|
| 10 |
+
"errors": "replace",
|
| 11 |
+
"extra_special_tokens": [
|
| 12 |
+
"[Q] ",
|
| 13 |
+
"[D] "
|
| 14 |
+
],
|
| 15 |
+
"image_token": "<|image_pad|>",
|
| 16 |
+
"is_local": true,
|
| 17 |
+
"model_max_length": 262144,
|
| 18 |
+
"model_specific_special_tokens": {
|
| 19 |
+
"audio_bos_token": "<|audio_start|>",
|
| 20 |
+
"audio_eos_token": "<|audio_end|>",
|
| 21 |
+
"audio_token": "<|audio_pad|>",
|
| 22 |
+
"image_token": "<|image_pad|>",
|
| 23 |
+
"video_token": "<|video_pad|>",
|
| 24 |
+
"vision_bos_token": "<|vision_start|>",
|
| 25 |
+
"vision_eos_token": "<|vision_end|>"
|
| 26 |
+
},
|
| 27 |
+
"pad_token": "<|endoftext|>",
|
| 28 |
+
"pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
|
| 29 |
+
"processor_class": "Qwen3VLProcessor",
|
| 30 |
+
"split_special_tokens": false,
|
| 31 |
+
"tokenizer_class": "TokenizersBackend",
|
| 32 |
+
"unk_token": null,
|
| 33 |
+
"video_token": "<|video_pad|>",
|
| 34 |
+
"vision_bos_token": "<|vision_start|>",
|
| 35 |
+
"vision_eos_token": "<|vision_end|>"
|
| 36 |
+
}
|