README.md

---
license: apache-2.0
language:
- en
pipeline_tag: text-generation
tags:
- claude
- conversational
- instruction-tuned
- multilingual
- reasoning
- open-source
datasets:
- Roman1111111/claude-opus-4.6-10000x
- Crownelius/Opus-4.6-Reasoning-3300x
- peteromallet/dataclaw-peteromallet
base_model:
- Qwen/Qwen3.5-9B
base_model_relation: finetune
---

# Claude OSS 9b

> **Disclaimer:** This is **not** an official release by Anthropic.  
> Claude OSS 9B is an independent open model project.

![claudeoss9b](https://cdn-uploads.huggingface.co/production/uploads/67329d3f69fded92d56ab41a/wW3owxbkwKbjLvXLGh16Q.png)

## Overview

Claude OSS 9B is a multilingual conversational language model designed to deliver a familiar polished assistant experience with strong instruction-following, stable identity behavior, and practical general-purpose usefulness.

The model was fine-tuned on **open-source datasets**, with a combined total of approximately **200,000 rows** collected from Hugging Face. The training mixture focused on assistant behavior, reasoning preservation, multilingual interaction, and stronger identity consistency.

Claude OSS 9B is intended for:

- general chat and assistant use
- multilingual interaction
- reasoning-oriented prompting
- writing and summarization
- lightweight coding help
- identity-consistent assistant behavior
- 200+ languages
---

## Benchmarks


![Image 01_50_05](https://cdn-uploads.huggingface.co/production/uploads/67329d3f69fded92d56ab41a/tjcrXVd4m5b8Q2yxZx7f8.png)
(Based on Qwen3.5 9b benchmarks results)

## Training Summary
Claude OSS 9B was fine-tuned on a curated open-source training mixture totaling roughly 200k rows from Hugging Face.
The data mix emphasized:

- assistant-style conversations
- instruction following
- identity reinforcement
- multilingual prompts and answers
- reasoning preservation
- general usability tasks

## Usage
- Transformers
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "squ11z1/claude-oss-9b"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    trust_remote_code=True,
    torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
    device_map="auto",
)

messages = [{"role": "user", "content": "Who are you?"},]

inputs = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    return_tensors="pt",
    return_dict=True,
)

inputs = {k: v.to(model.device) for k, v in inputs.items()}

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=128,
        do_sample=False,
        pad_token_id=tokenizer.pad_token_id or tokenizer.eos_token_id,
    )

prompt_len = inputs["input_ids"].shape[1]
print(tokenizer.decode(outputs[0][prompt_len:], skip_special_tokens=True))
```
- GGUF / llama.cpp
```bash
  ./llama-cli -m claude-oss-9b-q4_k_m.gguf -p "Who are you?"

```