Publish tested PR 3225 tiny MTP fixture

Toy-trained checkpoint and validation evidence for llm-compressor 2d52420. Cards document architecture-specific limitations.

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,79 @@
+---
+library_name: transformers
+license: apache-2.0
+tags:
+- llm-compressor
+- tiny-model
+- mtp
+- pr-3225
+- test-fixture
+---
+
+# Inkling-0.92B-MTP-PR3225
+
+Toy-trained BF16 source for [LLM Compressor PR #3225](https://github.com/vllm-project/llm-compressor/pull/3225). This is a small architecture and checkpoint test fixture, not a production language model. Its backbone was initialized randomly and trained on the repository tiny-model skill's toy text corpus; no pretrained base-model weights were used.
+
+## Model and provenance
+
+- Original architecture/tokenizer: [thinkingmachines/Inkling](https://huggingface.co/thinkingmachines/Inkling), revision `828496eeae4c243ff1a22f7f28ff83694f2f7bc9`.
+- Transformers class: `InklingForConditionalGeneration`.
+- Total source parameters including synthetic MTP: **915,998,469** (0.916B).
+- Backbone parameters: **830,199,043**; MTP parameters: **85,799,426**.
+- Estimated active backbone parameters: **783,013,123**. This routing estimate includes embeddings/dense components, excludes MTP, and is not a FLOP measurement.
+- Precision: BF16.
+- Checkpoint: 4 indexed safetensors shards; 109 indexed tensors. Tokenizer and processor assets are included.
+
+## Configuration
+
+| Field | Original | Tiny |
+| --- | --- | --- |
+| `num_hidden_layers` | 66 | 3 |
+| `hidden_size` | 6144 | 1536 |
+| `intermediate_size` | 3072 | 6144 |
+| `moe_intermediate_size` | not explicitly set | 1024 |
+| `n_routed_experts` | 256 | 16 |
+| `num_experts_per_tok` | 6 | 6 |
+| `num_attention_heads` | 64 | 16 |
+| `num_key_value_heads` | 8 | 8 |
+| `num_mtp_layers` | not explicitly set | 2 |
+
+The saved `config.json` is authoritative. Depth is reduced; the limitations below apply.
+Inkling retains sliding and full attention, dense and sparse MLPs, and its multimodal branches. Only text was trained/tested; image/audio quality was not evaluated.
+
+## Validation and scope
+
+The full multimodal model's text-only training and backbone reload passed; its nonuniform attention/MLP layout is retained. Single-process pipeline loading passed. The two-rank pipeline failed during fused-MoE/meta tensor handling, and vLLM inspection encountered the environment's NumPy/Numba compatibility error. No architecture or loader override was applied. MFPTQ, quantized serving, and MTP execution have not been validated for this fixture.
+
+The reloaded **BF16 backbone** toy perplexity was **1.570810** on the same small corpus used for training. This demonstrates learning/reload integrity, not generalization or benchmark quality.
+
+MTP projections are synthetic initializations and decoder weights were copied from trained backbone blocks. **MTP was not separately trained**, so these artifacts do not establish draft acceptance quality or inference acceleration.
+
+The tested model-based environment used **Transformers 5.17.0**, Torch 2.14.0+cu130, LLM Compressor [`2d52420`](https://github.com/vllm-project/llm-compressor/commit/2d5242056a8a00028686a303771ffc8e2fe27d07), and compressed-tensors [`e69c8dc`](https://github.com/vllm-project/compressed-tensors/commit/e69c8dc58aa152e5f5e36e85800ec1d2e7de5271). Serving smoke checks used vLLM 0.30.0. GLM/DeepSeek model-based MTP loading failed on Transformers **5.15.0**; **5.16 was not tested**. Calibration-dependent MTP quantization remains follow-up work.
+
+The model class/loader was not patched to make tests pass. Construction changes were configuration reductions and explicit synthetic checkpoint fixture creation. Structured results and configuration provenance are in [validation.json](validation.json); [artifact-manifest.json](artifact-manifest.json) records checkpoint file hashes.
+
+## Backbone loading
+
+With the tested Transformers environment (and compressed-tensors for FP8), this loads the backbone. MTP is opt-in; ordinary Transformers backbone generation does not execute it.
+
+```python
+import torch
+from transformers import InklingForConditionalGeneration, AutoTokenizer
+
+model_id = "inference-optimization/Inkling-0.92B-MTP-PR3225"
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+model = InklingForConditionalGeneration.from_pretrained(
+    model_id, dtype=torch.bfloat16, attn_implementation="eager",
+)
+inputs = tokenizer("The capital of France is", return_tensors="pt")
+output = model.generate(**inputs, max_new_tokens=16, do_sample=False)
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+```
+
+## Training
+
+Random seed 3225; AdamW with learning rate 0.0004 and weight decay 0.01; batch size 2; text truncated to 160 tokens. Training stopped after three consecutive checks at toy perplexity <=3. The corpus follows the [repository tiny-model workflow](https://github.com/vllm-project/llm-compressor/tree/2d5242056a8a00028686a303771ffc8e2fe27d07/.agents/skills/create-tiny-model), with one additional short saying. These are the trained/reloaded artifacts used in testing.
+
+## License and attribution
+
+Upstream architecture/tokenizer provenance and license: [thinkingmachines/Inkling](https://huggingface.co/thinkingmachines/Inkling/tree/828496eeae4c243ff1a22f7f28ff83694f2f7bc9).

artifact-manifest.json

@@ -0,0 +1,58 @@
+{
+  "repo_id": "inference-optimization/Inkling-0.92B-MTP-PR3225",
+  "files": {
+    "README.md": {
+      "size": 5061,
+      "sha256": "0e359db0aa89946ba67206a687d863822b260ced504331b5643ef516650acb11"
+    },
+    "chat_template.jinja": {
+      "size": 6294,
+      "sha256": "0aa1aa0c729d90176dcaa00c440c8faffca2957ffb2cc4b79456ee6d02bcf43b"
+    },
+    "config.json": {
+      "size": 2969,
+      "sha256": "17e1dad01f6c9770b10b94f7b60175327280c1bb7b71a49f43c1b187aba4ab36"
+    },
+    "generation_config.json": {
+      "size": 179,
+      "sha256": "122edeeea81ea41601f463a5f6fbe56a1d0e46b1ab2d6e1be2396785647ba4b5"
+    },
+    "model-00001-of-00003.safetensors": {
+      "size": 617545864,
+      "sha256": "9d916d3ca74f99a0b2f8bbc2f6e2322f8cc80a42b8a2c0e57c670220e57b2a77"
+    },
+    "model-00002-of-00003.safetensors": {
+      "size": 617545864,
+      "sha256": "2a067909dfe308248e1cf161128b86b30ff495833ad6fb8a5c4394d4996f0237"
+    },
+    "model-00003-of-00003.safetensors": {
+      "size": 425314046,
+      "sha256": "b8ef053454619d9d8cf42d79bbfa31c6430770e3c54224b32776ed14799d3d23"
+    },
+    "model-mtp.safetensors": {
+      "size": 171603828,
+      "sha256": "1c3b2574db8171c2d437d57b77e3fd164c8c1bea29f1e3c2938d6ab171ea1265"
+    },
+    "model.safetensors.index.json": {
+      "size": 8968,
+      "sha256": "9691ccd63bae634ecf34fcfcd0ac4418fad98532f3aa4f2d33d41e6feb8fa066"
+    },
+    "processor_config.json": {
+      "size": 1160,
+      "sha256": "3e10ba0f5b06b197bf194a2902ca108e0ffa0b895d14c40359f60e6a8964b752"
+    },
+    "tokenizer.json": {
+      "size": 27875796,
+      "sha256": "c470271fb7649eef922e660e757b5f076e2a94adc20df7fad7df85c46bc48eac"
+    },
+    "tokenizer_config.json": {
+      "size": 2011,
+      "sha256": "a5fac6a9ccffb5c681e2877cc749b612a8001dddcd65badaddeb00cadd1a750d"
+    },
+    "validation.json": {
+      "size": 15043,
+      "sha256": "ece68c2894ff25ffbe3450cf0ad9ca193f22196d9d9f10f3112ff008e6ee2eb7"
+    }
+  },
+  "weight_and_config_changes_during_publication": false
+}

chat_template.jinja

@@ -0,0 +1,129 @@
+{%- set effort_map = {"none": 0.0, "minimal": 0.1, "low": 0.2, "medium": 0.7, "high": 0.9, "max": 0.99} -%}
+{%- set role_token = {"user": "<|message_user|>", "assistant": "<|message_model|>", "system": "<|message_system|>", "tool": "<|message_tool|>"} -%}
+
+{%- macro emit_thinking_effort() -%}
+    {%- set eff = reasoning_effort if reasoning_effort is defined and reasoning_effort is not none else 0.9 -%}
+    {%- if eff is string -%}
+        {%- set key = eff | trim -%}
+        {%- if key not in effort_map -%}
+            {{- raise_exception("Unknown reasoning_effort: " ~ eff) -}}
+        {%- endif -%}
+        {%- set num = effort_map[key] -%}
+    {%- else -%}
+        {%- set num = eff | float -%}
+    {%- endif -%}
+    {%- if num < 0.0 or num > 0.99 -%}
+        {{- raise_exception("reasoning_effort must be in [0.0, 0.99]") -}}
+    {%- endif -%}
+    {{- "<|message_system|><|content_text|>Thinking effort level: " -}}
+    {%- if num == 0.0 -%}0{%- else -%}{{ num }}{%- endif -%}
+    {{- "<|end_message|>" -}}
+{%- endmacro -%}
+
+{%- if tools -%}
+    {%- set tool_state = namespace(specs=[]) -%}
+    {%- for tool in tools -%}
+        {%- set fn = tool.function if tool.function is defined else tool -%}
+        {%- set spec = {
+            "description": (fn.description if fn.description is defined and fn.description else ""),
+            "name": fn.name,
+            "parameters": (fn.parameters if fn.parameters is defined and fn.parameters else {}),
+            "type": (tool.type if tool.type is defined and tool.type else "function"),
+        } -%}
+        {%- set tool_state.specs = tool_state.specs + [spec] -%}
+    {%- endfor -%}
+    {{- "<|message_system|>tool_declare<|content_xml|>" -}}
+    {{- tool_state.specs | tojson(sort_keys=true, separators=(",", ":")) -}}
+    {{- "<|end_message|>" -}}
+{%- endif -%}
+
+{%- set state = namespace(effort_emitted=false) -%}
+{%- for message in messages -%}
+    {%- if message.role not in role_token -%}
+        {{- raise_exception("Unknown message role: " ~ message.role) -}}
+    {%- endif -%}
+    {%- if not state.effort_emitted and message.role != "system" -%}
+        {{- emit_thinking_effort() -}}
+        {%- set state.effort_emitted = true -%}
+    {%- endif -%}
+
+    {%- set rtok = role_token[message.role] -%}
+
+    {%- if message.role == "tool" -%}
+        {%- set tool_name_state = namespace(name="") -%}
+        {%- if message.name is defined and message.name -%}
+            {%- set tool_name_state.name = message.name -%}
+        {%- elif message.tool_call_id is defined and message.tool_call_id -%}
+            {%- for prev in messages -%}
+                {%- if prev.role == "assistant" and prev.tool_calls -%}
+                    {%- for tc in prev.tool_calls -%}
+                        {%- if tc.id is defined and tc.id == message.tool_call_id and tc.function.name is defined -%}
+                            {%- set tool_name_state.name = tc.function.name -%}
+                        {%- endif -%}
+                    {%- endfor -%}
+                {%- endif -%}
+            {%- endfor -%}
+        {%- endif -%}
+        {{- rtok -}}
+        {%- if tool_name_state.name -%}{{- tool_name_state.name -}}{%- endif -%}
+        {{- "<|content_text|>" -}}
+        {%- if message.content is string -%}{{- message.content -}}{%- endif -%}
+        {{- "<|end_message|>" -}}
+
+    {%- else -%}
+        {%- if message.role == "assistant" and message.reasoning_content is defined and message.reasoning_content -%}
+            {{- "<|message_model|><|content_thinking|>" ~ message.reasoning_content ~ "<|end_message|>" -}}
+        {%- endif -%}
+
+        {%- if message.content is string -%}
+            {{- rtok ~ "<|content_text|>" ~ message.content ~ "<|end_message|>" -}}
+        {%- elif message.content -%}
+            {%- for part in message.content -%}
+                {%- if part is string -%}
+                    {{- rtok ~ "<|content_text|>" ~ part ~ "<|end_message|>" -}}
+                {%- elif part.type is not defined or part.type in ("text", "input_text") -%}
+                    {%- set text_part = (part.text if part.text is defined and part.text is string else "") -%}
+                    {{- rtok ~ "<|content_text|>" ~ text_part ~ "<|end_message|>" -}}
+                {%- elif part.type in ("image", "input_image", "image_url") -%}
+                    {{- rtok ~ "<|content_image|><|unused_200054|><|end_message|>" -}}
+                {%- elif part.type in ("audio", "input_audio", "audio_url") -%}
+                    {{- rtok ~ "<|content_audio_input|><|unused_200053|><|audio_end|><|end_message|>" -}}
+                {%- else -%}
+                    {{- raise_exception("Unsupported content part type: " ~ part.type) -}}
+                {%- endif -%}
+            {%- endfor -%}
+        {%- endif -%}
+
+        {%- if message.role == "assistant" and message.tool_calls -%}
+            {%- for tc in message.tool_calls -%}
+                {%- set fn = tc.function -%}
+                {%- if fn.name is not defined or fn.name is not string -%}
+                    {{- raise_exception("tool call function name must be a string") -}}
+                {%- endif -%}
+                {%- set args = fn.arguments if fn.arguments is defined and fn.arguments else {} -%}
+                {%- if args is string -%}
+                    {{- raise_exception("tool call arguments must be a parsed object, not a JSON string; canonicalize upstream") -}}
+                {%- endif -%}
+                {%- if args is not mapping -%}
+                    {{- raise_exception("tool call arguments must be an object") -}}
+                {%- endif -%}
+                {{- "<|message_model|>" ~ fn.name ~ "<|content_invoke_tool_json|>" -}}
+                {{- '{"name":' ~ (fn.name | tojson(sort_keys=true, separators=(",", ":"))) ~ ',"args":' -}}
+                {{- (args | tojson(sort_keys=true, separators=(",", ":"))) -}}
+                {{- "}<|end_message|>" -}}
+            {%- endfor -%}
+        {%- endif -%}
+
+        {%- if message.role == "assistant" -%}
+            {{- "<|content_model_end_sampling|>" -}}
+        {%- endif -%}
+    {%- endif -%}
+{%- endfor -%}
+
+{%- if not state.effort_emitted -%}
+    {{- emit_thinking_effort() -}}
+{%- endif -%}
+
+{%- if add_generation_prompt -%}
+    {{- "<|message_model|>" -}}
+{%- endif -%}

config.json

@@ -0,0 +1,120 @@
+{
+  "architectures": [
+    "InklingForConditionalGeneration"
+  ],
+  "audio_bos_token_id": 200020,
+  "audio_config": {
+    "audio_mode": "dmel",
+    "bias": false,
+    "decoder_dmodel": 6144,
+    "dmel_max_value": 2.0,
+    "dmel_min_value": -7.0,
+    "dtype": "bfloat16",
+    "initializer_range": 0.02,
+    "mel_vocab_size": 16,
+    "model_type": "inkling_audio",
+    "n_mel_bins": 80,
+    "rms_norm_eps": 1e-06,
+    "text_hidden_size": 1536,
+    "use_audio_norm": true
+  },
+  "audio_token_id": 200053,
+  "dtype": "bfloat16",
+  "eos_token_id": 200006,
+  "image_bos_token_id": 200005,
+  "image_token_id": 200054,
+  "model_type": "inkling_mm_model",
+  "mtp_config": {
+    "chain_hidden_post_norm": false,
+    "local_layer_ids": [
+      0
+    ],
+    "num_nextn_predict_layers": 2
+  },
+  "pad_token_id": 200006,
+  "text_config": {
+    "attention_dropout": 0.0,
+    "bos_token_id": 1,
+    "chain_hidden_post_norm": false,
+    "conv_kernel_size": 4,
+    "d_rel": 16,
+    "dtype": "bfloat16",
+    "eos_token_id": 2,
+    "final_logit_softcapping": null,
+    "gate_activation": "sigmoid",
+    "head_dim": 128,
+    "hidden_act": "silu",
+    "hidden_size": 1536,
+    "initializer_range": 0.02,
+    "intermediate_size": 6144,
+    "layer_types": [
+      "hybrid_sliding",
+      "hybrid_sliding",
+      "hybrid"
+    ],
+    "local_layer_ids": [
+      0,
+      1
+    ],
+    "log_scaling_alpha": 0.1,
+    "log_scaling_n_floor": 128000,
+    "logits_mup_width_multiplier": 6.0,
+    "max_position_embeddings": 1048576,
+    "mlp_layer_types": [
+      "dense",
+      "dense",
+      "sparse"
+    ],
+    "model_type": "inkling_text",
+    "moe_intermediate_size": 1024,
+    "mtp_hidden_states_first": true,
+    "mtp_local_layer_ids": [
+      0
+    ],
+    "n_routed_experts": 16,
+    "n_shared_experts": 2,
+    "norm_after_topk": true,
+    "num_attention_heads": 16,
+    "num_experts_per_tok": 6,
+    "num_hidden_layers": 3,
+    "num_key_value_heads": 8,
+    "num_mtp_layers": 2,
+    "number_of_conv_states": 4,
+    "o_bias": false,
+    "pad_token_id": null,
+    "q_bias": false,
+    "rel_extent": 1024,
+    "rms_norm_eps": 1e-06,
+    "rms_norm_eps_moe_gate": 1e-06,
+    "route_scale": 8.0,
+    "shared_expert_sink": true,
+    "sliding_window_size": 512,
+    "swa_head_dim": 128,
+    "swa_num_attention_heads": 16,
+    "swa_num_key_value_heads": 8,
+    "unpadded_vocab_size": 200058,
+    "use_embed_norm": true,
+    "use_gate_bias": true,
+    "use_global_scale": true,
+    "use_sconv": true,
+    "vocab_size": 201024
+  },
+  "transformers_version": "5.17.0",
+  "vision_config": {
+    "decoder_dmodel": 6144,
+    "dtype": "bfloat16",
+    "hidden_size": 1024,
+    "initializer_range": 0.02,
+    "model_type": "inkling_vision",
+    "n_channels": 3,
+    "n_layers": 4,
+    "num_attention_heads": 16,
+    "num_channels": 3,
+    "patch_size": 40,
+    "rms_norm_eps": 1e-06,
+    "temporal_patch_size": 2,
+    "text_hidden_size": 1536,
+    "use_vision_norm": true,
+    "vision_encoder_type": "hmlp"
+  }
+}

generation_config.json

@@ -0,0 +1,8 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 200006,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "transformers_version": "5.17.0"
+}

model-00001-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d916d3ca74f99a0b2f8bbc2f6e2322f8cc80a42b8a2c0e57c670220e57b2a77
+size 617545864

model-00002-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a067909dfe308248e1cf161128b86b30ff495833ad6fb8a5c4394d4996f0237
+size 617545864

model-00003-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8ef053454619d9d8cf42d79bbfa31c6430770e3c54224b32776ed14799d3d23
+size 425314046

model-mtp.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c3b2574db8171c2d437d57b77e3fd164c8c1bea29f1e3c2938d6ab171ea1265
+size 171603828

model.safetensors.index.json

@@ -0,0 +1,116 @@
+{
+  "metadata": {
+    "total_size": 1832009602
+  },
+  "weight_map": {
+    "model.audio.encoder.weight": "model-00003-of-00003.safetensors",
+    "model.audio.final_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.embed.weight": "model-00002-of-00003.safetensors",
+    "model.llm.embed_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.attn.k_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.attn.k_sconv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.attn.q_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.attn.rel_logits_proj.proj": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.attn.v_sconv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.attn.wk_dv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.attn.wo_ud.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.attn.wq_du.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.attn.wr_du.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.attn.wv_dv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.attn_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.attn_sconv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.mlp.global_scale": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.mlp.w13_dn.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.mlp.w2_md.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.mlp_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.0.mlp_sconv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.attn.k_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.attn.k_sconv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.attn.q_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.attn.rel_logits_proj.proj": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.attn.v_sconv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.attn.wk_dv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.attn.wo_ud.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.attn.wq_du.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.attn.wr_du.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.attn.wv_dv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.attn_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.attn_sconv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.mlp.global_scale": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.mlp.w13_dn.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.mlp.w2_md.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.mlp_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.1.mlp_sconv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.attn.k_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.attn.k_sconv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.attn.q_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.attn.rel_logits_proj.proj": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.attn.v_sconv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.attn.wk_dv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.attn.wo_ud.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.attn.wq_du.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.attn.wr_du.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.attn.wv_dv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.attn_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.attn_sconv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.mlp.experts.w13_weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.mlp.experts.w2_weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.mlp.gate.bias": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.mlp.gate.global_scale": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.mlp.gate.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.mlp.shared_experts.shared_w13_weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.mlp.shared_experts.shared_w2_weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.mlp_norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.layers.2.mlp_sconv.weight": "model-00003-of-00003.safetensors",
+    "model.llm.norm.weight": "model-00003-of-00003.safetensors",
+    "model.llm.unembed.weight": "model-00001-of-00003.safetensors",
+    "model.mtp.layers.0.embed_norm.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.hidden_norm.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.input_proj.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.attn.k_norm.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.attn.k_sconv.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.attn.q_norm.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.attn.rel_logits_proj.proj": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.attn.v_sconv.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.attn.wk_dv.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.attn.wo_ud.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.attn.wq_du.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.attn.wr_du.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.attn.wv_dv.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.attn_norm.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.attn_sconv.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.mlp.global_scale": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.mlp.w13_dn.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.mlp.w2_md.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.mlp_norm.weight": "model-mtp.safetensors",
+    "model.mtp.layers.0.transformer_block.mlp_sconv.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.embed_norm.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.hidden_norm.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.input_proj.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.attn.k_norm.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.attn.k_sconv.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.attn.q_norm.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.attn.rel_logits_proj.proj": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.attn.v_sconv.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.attn.wk_dv.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.attn.wo_ud.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.attn.wq_du.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.attn.wr_du.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.attn.wv_dv.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.attn_norm.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.attn_sconv.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.mlp.global_scale": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.mlp.w13_dn.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.mlp.w2_md.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.mlp_norm.weight": "model-mtp.safetensors",
+    "model.mtp.layers.1.transformer_block.mlp_sconv.weight": "model-mtp.safetensors",
+    "model.visual.final_norm.weight": "model-00003-of-00003.safetensors",
+    "model.visual.layers.linear_0.weight": "model-00003-of-00003.safetensors",
+    "model.visual.layers.linear_1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.layers.linear_2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.layers.linear_3.weight": "model-00003-of-00003.safetensors",
+    "model.visual.layers.norm_0.weight": "model-00003-of-00003.safetensors",
+    "model.visual.layers.norm_1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.layers.norm_2.weight": "model-00003-of-00003.safetensors"
+  }
+}

processor_config.json

@@ -0,0 +1,47 @@
+{
+  "audio_bos_token": "<|content_audio_input|>",
+  "audio_token": "<|unused_200053|>",
+  "dmel_max_value": 2.0,
+  "dmel_min_value": -7.0,
+  "feature_extractor": {
+    "audio_token_duration_s": 0.05,
+    "feature_extractor_type": "InklingFeatureExtractor",
+    "feature_size": 80,
+    "hop_length": 800,
+    "n_fft": 1600,
+    "padding_side": "right",
+    "padding_value": 0.0,
+    "return_attention_mask": true,
+    "sampling_rate": 16000,
+    "window_size": 1600,
+    "window_size_multiplier": 2.0
+  },
+  "image_bos_token": "<|content_image|>",
+  "image_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_mean": [
+      0.48145466,
+      0.4578275,
+      0.40821073
+    ],
+    "image_processor_type": "InklingImageProcessor",
+    "image_std": [
+      0.26862954,
+      0.26130258,
+      0.27577711
+    ],
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "rescale_image_max_upscaled_long_edge": 2048,
+    "size": {
+      "height": 40,
+      "width": 40
+    }
+  },
+  "image_token": "<|unused_200054|>",
+  "num_dmel_bins": 16,
+  "processor_class": "InklingProcessor"
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c470271fb7649eef922e660e757b5f076e2a94adc20df7fad7df85c46bc48eac
+size 27875796

tokenizer_config.json

@@ -0,0 +1,50 @@
+{
+  "audio": "<|audio|>",
+  "audio_end": "<|audio_end|>",
+  "backend": "tokenizers",
+  "begin_of_text": "<|begin_of_text|>",
+  "clean_up_tokenization_spaces": false,
+  "content_audio_input": "<|content_audio_input|>",
+  "content_image": "<|content_image|>",
+  "content_invoke_tool_json": "<|content_invoke_tool_json|>",
+  "content_invoke_tool_text": "<|content_invoke_tool_text|>",
+  "content_model_end_sampling": "<|content_model_end_sampling|>",
+  "content_text": "<|content_text|>",
+  "content_thinking": "<|content_thinking|>",
+  "content_tool_error": "<|content_tool_error|>",
+  "content_xml": "<|content_xml|>",
+  "end_message": "<|end_message|>",
+  "endoftext": "<|endoftext|>",
+  "eos_token": "<|content_model_end_sampling|>",
+  "fix_mistral_regex": false,
+  "is_local": true,
+  "local_files_only": false,
+  "message_model": "<|message_model|>",
+  "message_system": "<|message_system|>",
+  "message_tool": "<|message_tool|>",
+  "message_user": "<|message_user|>",
+  "model_max_length": 1000000000000000019884624838656,
+  "model_specific_special_tokens": {
+    "audio": "<|audio|>",
+    "audio_end": "<|audio_end|>",
+    "begin_of_text": "<|begin_of_text|>",
+    "content_audio_input": "<|content_audio_input|>",
+    "content_image": "<|content_image|>",
+    "content_invoke_tool_json": "<|content_invoke_tool_json|>",
+    "content_invoke_tool_text": "<|content_invoke_tool_text|>",
+    "content_model_end_sampling": "<|content_model_end_sampling|>",
+    "content_text": "<|content_text|>",
+    "content_thinking": "<|content_thinking|>",
+    "content_tool_error": "<|content_tool_error|>",
+    "content_xml": "<|content_xml|>",
+    "end_message": "<|end_message|>",
+    "endoftext": "<|endoftext|>",
+    "message_model": "<|message_model|>",
+    "message_system": "<|message_system|>",
+    "message_tool": "<|message_tool|>",
+    "message_user": "<|message_user|>"
+  },
+  "pad_token": "<|content_model_end_sampling|>",
+  "processor_class": "InklingProcessor",
+  "tokenizer_class": "TokenizersBackend"
+}

validation.json

@@ -0,0 +1,549 @@
+{
+  "llm_compressor_commit": "2d5242056a8a00028686a303771ffc8e2fe27d07",
+  "compressed_tensors_commit": "e69c8dc58aa152e5f5e36e85800ec1d2e7de5271",
+  "transformers": "5.17.0",
+  "torch": "2.14.0+cu130",
+  "vllm": "0.30.0",
+  "base_architecture": "thinkingmachines/Inkling",
+  "base_revision": "828496eeae4c243ff1a22f7f28ff83694f2f7bc9",
+  "fixture": "inkling",
+  "variant": "source",
+  "parameter_counts": {
+    "architecture": "InklingForConditionalGeneration",
+    "layers_only_parameters": 62689820678,
+    "backbone_parameters": 830199043,
+    "mtp_parameters": 85799426,
+    "total_parameters": 915998469,
+    "changes": {
+      "dtype": {
+        "original": null,
+        "tiny": "bfloat16"
+      },
+      "text_config": {
+        "original": {
+          "architectures": null,
+          "output_hidden_states": false,
+          "return_dict": true,
+          "dtype": "bfloat16",
+          "chunk_size_feed_forward": 0,
+          "is_encoder_decoder": false,
+          "id2label": {
+            "0": "LABEL_0",
+            "1": "LABEL_1"
+          },
+          "label2id": {
+            "LABEL_0": 0,
+            "LABEL_1": 1
+          },
+          "problem_type": null,
+          "vocab_size": 201024,
+          "unpadded_vocab_size": 200058,
+          "hidden_size": 6144,
+          "num_hidden_layers": 66,
+          "num_attention_heads": 64,
+          "num_key_value_heads": 8,
+          "head_dim": 128,
+          "swa_num_attention_heads": 64,
+          "swa_num_key_value_heads": 16,
+          "swa_head_dim": 128,
+          "sliding_window_size": 512,
+          "d_rel": 16,
+          "rel_extent": 1024,
+          "log_scaling_n_floor": 128000,
+          "log_scaling_alpha": 0.1,
+          "local_layer_ids": [
+            0,
+            1,
+            2,
+            3,
+            4,
+            6,
+            7,
+            8,
+            9,
+            10,
+            12,
+            13,
+            14,
+            15,
+            16,
+            18,
+            19,
+            20,
+            21,
+            22,
+            24,
+            25,
+            26,
+            27,
+            28,
+            30,
+            31,
+            32,
+            33,
+            34,
+            36,
+            37,
+            38,
+            39,
+            40,
+            42,
+            43,
+            44,
+            45,
+            46,
+            48,
+            49,
+            50,
+            51,
+            52,
+            54,
+            55,
+            56,
+            57,
+            58,
+            60,
+            61,
+            62,
+            63,
+            64
+          ],
+          "layer_types": [
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid"
+          ],
+          "max_position_embeddings": 1048576,
+          "rms_norm_eps": 1e-06,
+          "conv_kernel_size": 4,
+          "mlp_layer_types": [
+            "dense",
+            "dense",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse",
+            "sparse"
+          ],
+          "intermediate_size": 24576,
+          "hidden_act": "silu",
+          "moe_intermediate_size": 3072,
+          "n_routed_experts": 256,
+          "num_experts_per_tok": 6,
+          "n_shared_experts": 2,
+          "shared_expert_sink": true,
+          "route_scale": 8.0,
+          "logits_mup_width_multiplier": 24.0,
+          "rms_norm_eps_moe_gate": 1e-06,
+          "attention_dropout": 0.0,
+          "initializer_range": 0.02,
+          "pad_token_id": null,
+          "bos_token_id": 1,
+          "eos_token_id": 2,
+          "num_mtp_layers": 8,
+          "chain_hidden_post_norm": false,
+          "mtp_hidden_states_first": true,
+          "mtp_local_layer_ids": [
+            0,
+            2,
+            4,
+            5,
+            6,
+            7
+          ],
+          "number_of_conv_states": 4,
+          "_name_or_path": "",
+          "q_bias": false,
+          "o_bias": false,
+          "use_embed_norm": true,
+          "use_sconv": true,
+          "final_logit_softcapping": null,
+          "use_gate_bias": true,
+          "gate_activation": "sigmoid",
+          "norm_after_topk": true,
+          "use_global_scale": true,
+          "model_type": "inkling_text",
+          "output_attentions": false
+        },
+        "tiny": {
+          "architectures": null,
+          "output_hidden_states": false,
+          "return_dict": true,
+          "dtype": "bfloat16",
+          "chunk_size_feed_forward": 0,
+          "is_encoder_decoder": false,
+          "id2label": {
+            "0": "LABEL_0",
+            "1": "LABEL_1"
+          },
+          "label2id": {
+            "LABEL_0": 0,
+            "LABEL_1": 1
+          },
+          "problem_type": null,
+          "vocab_size": 201024,
+          "unpadded_vocab_size": 200058,
+          "hidden_size": 1536,
+          "num_hidden_layers": 3,
+          "num_attention_heads": 16,
+          "num_key_value_heads": 8,
+          "head_dim": 128,
+          "swa_num_attention_heads": 16,
+          "swa_num_key_value_heads": 8,
+          "swa_head_dim": 128,
+          "sliding_window_size": 512,
+          "d_rel": 16,
+          "rel_extent": 1024,
+          "log_scaling_n_floor": 128000,
+          "log_scaling_alpha": 0.1,
+          "local_layer_ids": [
+            0,
+            1
+          ],
+          "layer_types": [
+            "hybrid_sliding",
+            "hybrid_sliding",
+            "hybrid"
+          ],
+          "max_position_embeddings": 1048576,
+          "rms_norm_eps": 1e-06,
+          "conv_kernel_size": 4,
+          "mlp_layer_types": [
+            "dense",
+            "dense",
+            "sparse"
+          ],
+          "intermediate_size": 6144,
+          "hidden_act": "silu",
+          "moe_intermediate_size": 1024,
+          "n_routed_experts": 16,
+          "num_experts_per_tok": 6,
+          "n_shared_experts": 2,
+          "shared_expert_sink": true,
+          "route_scale": 8.0,
+          "logits_mup_width_multiplier": 6.0,
+          "rms_norm_eps_moe_gate": 1e-06,
+          "attention_dropout": 0.0,
+          "initializer_range": 0.02,
+          "pad_token_id": null,
+          "bos_token_id": 1,
+          "eos_token_id": 2,
+          "num_mtp_layers": 2,
+          "chain_hidden_post_norm": false,
+          "mtp_hidden_states_first": true,
+          "mtp_local_layer_ids": [
+            0
+          ],
+          "number_of_conv_states": 4,
+          "_name_or_path": "",
+          "q_bias": false,
+          "o_bias": false,
+          "use_embed_norm": true,
+          "use_sconv": true,
+          "final_logit_softcapping": null,
+          "use_gate_bias": true,
+          "gate_activation": "sigmoid",
+          "norm_after_topk": true,
+          "use_global_scale": true,
+          "model_type": "inkling_text",
+          "output_attentions": false
+        }
+      },
+      "audio_config": {
+        "original": {
+          "architectures": null,
+          "output_hidden_states": false,
+          "return_dict": true,
+          "dtype": null,
+          "chunk_size_feed_forward": 0,
+          "is_encoder_decoder": false,
+          "id2label": {
+            "0": "LABEL_0",
+            "1": "LABEL_1"
+          },
+          "label2id": {
+            "LABEL_0": 0,
+            "LABEL_1": 1
+          },
+          "problem_type": null,
+          "n_mel_bins": 80,
+          "mel_vocab_size": 16,
+          "text_hidden_size": 6144,
+          "rms_norm_eps": 1e-06,
+          "initializer_range": 0.02,
+          "_name_or_path": "",
+          "decoder_dmodel": 6144,
+          "bias": false,
+          "dmel_min_value": -7.0,
+          "dmel_max_value": 2.0,
+          "use_audio_norm": true,
+          "audio_mode": "dmel",
+          "model_type": "inkling_audio",
+          "output_attentions": false
+        },
+        "tiny": {
+          "architectures": null,
+          "output_hidden_states": false,
+          "return_dict": true,
+          "dtype": null,
+          "chunk_size_feed_forward": 0,
+          "is_encoder_decoder": false,
+          "id2label": {
+            "0": "LABEL_0",
+            "1": "LABEL_1"
+          },
+          "label2id": {
+            "LABEL_0": 0,
+            "LABEL_1": 1
+          },
+          "problem_type": null,
+          "n_mel_bins": 80,
+          "mel_vocab_size": 16,
+          "text_hidden_size": 1536,
+          "rms_norm_eps": 1e-06,
+          "initializer_range": 0.02,
+          "_name_or_path": "",
+          "decoder_dmodel": 6144,
+          "bias": false,
+          "dmel_min_value": -7.0,
+          "dmel_max_value": 2.0,
+          "use_audio_norm": true,
+          "audio_mode": "dmel",
+          "model_type": "inkling_audio",
+          "output_attentions": false
+        }
+      },
+      "vision_config": {
+        "original": {
+          "architectures": null,
+          "output_hidden_states": false,
+          "return_dict": true,
+          "dtype": null,
+          "chunk_size_feed_forward": 0,
+          "is_encoder_decoder": false,
+          "id2label": {
+            "0": "LABEL_0",
+            "1": "LABEL_1"
+          },
+          "label2id": {
+            "LABEL_0": 0,
+            "LABEL_1": 1
+          },
+          "problem_type": null,
+          "text_hidden_size": 6144,
+          "patch_size": 40,
+          "temporal_patch_size": 2,
+          "num_channels": 3,
+          "hidden_size": 1024,
+          "n_layers": 4,
+          "num_attention_heads": 16,
+          "rms_norm_eps": 1e-06,
+          "initializer_range": 0.02,
+          "_name_or_path": "",
+          "vision_encoder_type": "hmlp",
+          "decoder_dmodel": 6144,
+          "n_channels": 3,
+          "use_vision_norm": true,
+          "model_type": "inkling_vision",
+          "output_attentions": false
+        },
+        "tiny": {
+          "architectures": null,
+          "output_hidden_states": false,
+          "return_dict": true,
+          "dtype": null,
+          "chunk_size_feed_forward": 0,
+          "is_encoder_decoder": false,
+          "id2label": {
+            "0": "LABEL_0",
+            "1": "LABEL_1"
+          },
+          "label2id": {
+            "LABEL_0": 0,
+            "LABEL_1": 1
+          },
+          "problem_type": null,
+          "text_hidden_size": 1536,
+          "patch_size": 40,
+          "temporal_patch_size": 2,
+          "num_channels": 3,
+          "hidden_size": 1024,
+          "n_layers": 4,
+          "num_attention_heads": 16,
+          "rms_norm_eps": 1e-06,
+          "initializer_range": 0.02,
+          "_name_or_path": "",
+          "vision_encoder_type": "hmlp",
+          "decoder_dmodel": 6144,
+          "n_channels": 3,
+          "use_vision_norm": true,
+          "model_type": "inkling_vision",
+          "output_attentions": false
+        }
+      },
+      "mtp_config": {
+        "original": {
+          "num_nextn_predict_layers": 8,
+          "chain_hidden_post_norm": false,
+          "local_layer_ids": [
+            0,
+            2,
+            4,
+            5,
+            6,
+            7
+          ]
+        },
+        "tiny": {
+          "num_nextn_predict_layers": 2,
+          "chain_hidden_post_norm": false,
+          "local_layer_ids": [
+            0
+          ]
+        }
+      }
+    }
+  },
+  "backbone_validation": {
+    "name": "inkling",
+    "passed": true,
+    "reloaded_perplexity": 1.570810437713275,
+    "model_patches": false,
+    "backbone_parameters": 830199043,
+    "backbone_activated_parameters_estimate": 783013123,
+    "generation": "The capital of France is Paris. Paris is the largest city in France and serves as the country's political,"
+  },
+  "mtp_training": "synthetic initialized projections with trained final decoder block copied; not separately trained",
+  "training": {
+    "seed": 3225,
+    "initialization": "random",
+    "optimizer": "AdamW",
+    "learning_rate": 0.0004,
+    "weight_decay": 0.01,
+    "batch_size": 2,
+    "max_sequence_length": 160,
+    "evaluation": "same toy corpus as training"
+  }
+}