README.md

---
language:
- en
license: mit
library_name: transformers
base_model: Qwen/Qwen2.5-Coder-7B-Instruct
pipeline_tag: question-answering
tags:
- isaac-sim
- omniverse
- robotics
- nvidia
- question-answering
- chat
---

![image/png](https://cdn-uploads.huggingface.co/production/uploads/67f44bfc74216a95759e2aa5/A2r-SoYBlK71CbjgGhH-J.png)


# Qwen2.5‑Coder‑7B‑Instruct‑Omni1.0 (Isaac Sim Assistant)

Purpose‑built coding assistant for NVIDIA Isaac Sim 5.0+ and Omniverse Kit 107.x. Fine‑tuned to deliver high-level assistance and troubleshooting help.

- Base: Qwen2.5‑Coder‑7B‑Instruct
- Repo: `TomBombadyl/Qwen2.5-Coder-7B-Instruct-Omni1.0`
- Interface: Chat messages or single‑turn text

---

## 0. Changelog
- 2025‑08‑13: Public Inference Endpoint live; added quickstart and examples.

---

## 1. Model Introduction

This model specializes in:
- Isaac Sim API usage and best practices
- Robot + sensor setup, physics, and extension patterns
- Robotics code generation and refactoring
- Diagnosing common Isaac Sim errors and warnings

Key features
- Chat‑optimized with structure‑aware prompting
- Good defaults for coding (stable, low randomness)
- Works via HTTP; no SDK required

---

## 2. Model Summary

- Architecture: Transformer (7B)
- Context: typical 4K+ (prompt truncation handled internally)
- Input formats: Chat `messages[]` or single‑turn `inputs`
- Output: `generated_text` plus simple token accounting

---

## 3. Try it now (Public Endpoint)

Current live URL (may change if redeployed):  
`https://k6yeljf74w9gw134.us-east4.gcp.endpoints.huggingface.cloud`

Recommended defaults for coding:
- temperature: 0.2
- top_p: 0.7
- max_new_tokens: 256 (raise as needed)

cURL (chat)
```bash
curl -s -X POST "https://k6yeljf74w9gw134.us-east4.gcp.endpoints.huggingface.cloud" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {"role":"system","content":"You are a helpful coding assistant for NVIDIA Isaac Sim."},
      {"role":"user","content":"Create a minimal script to spawn a URDF and enable PhysX."}
    ],
    "parameters": {"max_new_tokens":256, "temperature":0.2, "top_p":0.9}
  }'
```

Python
```python
import requests

url = "https://k6yeljf74w9gw134.us-east4.gcp.endpoints.huggingface.cloud"
payload = {
  "messages": [
    {"role":"system","content":"You are a helpful coding assistant for NVIDIA Isaac Sim."},
    {"role":"user","content":"How do I attach a camera sensor to a robot link?"}
  ],
  "parameters": {"max_new_tokens": 256, "temperature": 0.2, "top_p": 0.9}
}
print(requests.post(url, json=payload).json())
```

Single‑turn (non‑chat)
```bash
curl -s -X POST "https://k6yeljf74w9gw134.us-east4.gcp.endpoints.huggingface.cloud" \
  -H "Content-Type: application/json" \
  -d '{"inputs":"Say hello in one sentence.","parameters":{"max_new_tokens":64}}'
```

---

## 4. Inputs / Outputs

Inputs (choose one)
- Chat: `messages` as a list of `{role, content}`
- Single‑turn: `inputs` as a string

Common parameters
- `max_new_tokens`, `temperature`, `top_p`, `top_k`, `repetition_penalty`, `num_beams`, `do_sample`, `seed`
- `stop` (string or list)
- `max_input_tokens` (truncate prompt to reserve room for generation)

Response shape
```json
{
  "generated_text": "...",
  "input_tokens": 123,
  "generated_tokens": 256,
  "total_tokens": 379,
  "params": { "max_new_tokens": 256, "temperature": 0.2 }
}
```

---

## 5. Local usage (Transformers)

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

tok = AutoTokenizer.from_pretrained(
  "TomBombadyl/Qwen2.5-Coder-7B-Instruct-Omni1.0", trust_remote_code=True
)
model = AutoModelForCausalLM.from_pretrained(
  "TomBombadyl/Qwen2.5-Coder-7B-Instruct-Omni1.0",
  trust_remote_code=True, torch_dtype="auto", device_map="auto"
)

messages = [
  {"role":"system","content":"You are a helpful coding assistant for NVIDIA Isaac Sim."},
  {"role":"user","content":"Example: spawn a robot and start the simulation loop."}
]
prompt = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tok(prompt, return_tensors="pt").to(model.device)
out = model.generate(**inputs, max_new_tokens=256, temperature=0.2)
print(tok.decode(out[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))
```

---

## 6. Limitations

- May produce version‑specific code; verify imports and extension names for your Isaac Sim version.
- Not a substitute for official safety or hardware guidance.

---

## 7. License

MIT (see LICENSE).

---

## 8. Citation

```bibtex
@misc{qwen25-coder-isaac-sim,
  title  = {Qwen2.5-Coder-7B-Instruct-Omni1.0: Fine-tuned for NVIDIA Isaac Sim Development},
  author = {TomBombadyl},
  year   = {2024},
  url    = {https://huggingface.co/TomBombadyl/Qwen2.5-Coder-7B-Instruct-Omni1.0}
}
```

---

## 9. Contact

Open a Discussion on this model page with questions or feedback.