Instructions to use serda-dev/mamba-130m-hf-turkish with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use serda-dev/mamba-130m-hf-turkish with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="serda-dev/mamba-130m-hf-turkish")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("serda-dev/mamba-130m-hf-turkish")
model = AutoModelForCausalLM.from_pretrained("serda-dev/mamba-130m-hf-turkish", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use serda-dev/mamba-130m-hf-turkish with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "serda-dev/mamba-130m-hf-turkish"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "serda-dev/mamba-130m-hf-turkish",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/serda-dev/mamba-130m-hf-turkish

SGLang

How to use serda-dev/mamba-130m-hf-turkish with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "serda-dev/mamba-130m-hf-turkish" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "serda-dev/mamba-130m-hf-turkish",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "serda-dev/mamba-130m-hf-turkish" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "serda-dev/mamba-130m-hf-turkish",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use serda-dev/mamba-130m-hf-turkish with Docker Model Runner:
```
docker model run hf.co/serda-dev/mamba-130m-hf-turkish
```

serda-dev commited on Feb 13

Commit

9bee494

verified ·

1 Parent(s): 98057ae

Update README.md

Browse files

Files changed (1) hide show

README.md +20 -16

README.md CHANGED Viewed

@@ -16,27 +16,31 @@ tags:
 - continued-pretraining
 ---
-## Notice / Announcement (13.02.2026 11PM (UTC+3))
-Please read this notice regarding a few known issues affecting **both** `mamba-130m-hf-turkish` and `mamba-370m-hf-turkish`.
-This announcement is **identical for both repositories**, so you do **not** need to check the other model’s repo separately.
-### Known issue: text generation behavior (130M)
-- Due to an **embedding-related incompatibility** in `mamba-130m-hf-turkish`, the current **text generation** functionality may behave **buggily** (unstable or incorrect outputs).
-- This issue has been **fixed** in `mamba-370m-hf-turkish`, and we will apply the same fix to `mamba-130m-hf-turkish` **as soon as possible**.
-### Current model quality and roadmap
-- Despite the dataset limitations, both models generally produce **good Turkish surface form** (usage patterns, grammar alignment, and fluency).
-- However, there are still **logical / contextual consistency** issues (reasoning coherence, long-range consistency, factual reliability, etc.).
-- We will **keep the current approach** and continue improving the dataset pipeline (ongoing web scraping + cleaning).
-- When `mamba-2.8b-hf-turkish` is ready, we plan to **retrain and re-release** the full set of Turkish checkpoints together using the improved dataset.
-### In the meantime
-- Until that release, we recommend using these models primarily by **fine-tuning** them for your specific tasks.
-- Please don’t hesitate to **report additional issues** (generation bugs, tokenizer/embedding mismatches, edge cases, reproducibility problems, etc.).
----
----
 ---
 # Turkish Continued Pretraining of `mamba-130m-hf`

 - continued-pretraining
 ---
+## ⚠️ Notice / Duyuru (applies to both repos / iki repo için geçerli)
+#### (13.02.2026 11PM (UTC+3))
+> **EN:** This announcement is identical for **`mamba-130m-hf-turkish`** and **`mamba-370m-hf-turkish`** — you don’t need to check the other repository separately.
+> **TR:** Bu duyuru **`mamba-130m-hf-turkish`** ve **`mamba-370m-hf-turkish`** için aynıdır — diğer repoyu ayrıca kontrol etmenize gerek yoktur.
+<details>
+<summary><b>EN — Details</b></summary>
+Due to an **embedding-related incompatibility** in `mamba-130m-hf-turkish`, the current **text generation** behavior may be **buggy** (unstable or inconsistent outputs). This issue has been **fixed** in `mamba-370m-hf-turkish`, and we will port the same fix to `mamba-130m-hf-turkish` **as soon as possible**.
+Overall, Turkish fluency and grammar are generally solid, but **logical/contextual consistency** issues remain because of current dataset limitations. We are continuing to improve the dataset pipeline (ongoing web scraping and cleaning). When `mamba-2.8b-hf-turkish` is ready, we plan to **retrain and re-release** the Turkish checkpoints together using the improved dataset. Until then, we recommend using these models mainly via **fine-tuning**, and we appreciate any additional bug reports.
+</details>
+<details>
+<summary><b>TR — Detaylar</b></summary>
+`mamba-130m-hf-turkish` modelinde **embedding tarafındaki bir uyumsuzluk** nedeniyle mevcut **text generation** davranışı zaman zaman **buglu** çalışabiliyor (çıktılar tutarsız/kararsız olabiliyor). Bu sorun `mamba-370m-hf-turkish` modelinde **çözüldü** ve aynı düzeltmeyi `mamba-130m-hf-turkish` reposuna da **en kısa sürede** aktaracağız.
+Genel olarak Türkçe akıcılık ve gramer tarafı iyi; ancak mevcut dataset kısıtları nedeniyle **mantıksal bağlam ve tutarlılık** problemleri hâlâ görülebilir. Dataset hattını (web scrape + temizlik) iyileştirmeye devam ediyoruz. `mamba-2.8b-hf-turkish` hazır olduğunda, geliştirilmiş dataset ile Türkçe checkpoint’leri **birlikte yeniden eğitip yeniden yayınlamayı** planlıyoruz. O zamana kadar modelleri ağırlıklı olarak **fine-tune ederek** kullanmanızı öneririz; ek hataları bildirmekten çekinmeyin.
+</details>
 ---
 # Turkish Continued Pretraining of `mamba-130m-hf`