Skip to main content

Qwen 2.5 Coder 7B

Code-specialized LLM for generation, completion, debugging, and explanation.

Overview​

  • model_name: qwen-2.5-coder-7b (curated base)
  • Source: Qwen/Qwen2.5-Coder-7B-Instruct
  • Scale: ~7B parameters
  • Served on: GPU via vLLM (transformers fallback)
  • Public endpoint: qwen-2.5-coder-7b-public (task type code, billed per request in USD)

When to use​

✅ Code generation across dozens of languages, code completion, bug fixing, refactoring, test generation, and code explanation.

❌ For general chat use Mistral 7B; for non-technical content pick a general model.

Live specs​

Context window, VRAM footprint, price, and readiness come from the live catalog — this page does not hardcode them:

curl "https://api.colabhive.com/api/builder/v1/endpoints?visibility=public&search=qwen-2.5-coder"

Quick start​

from colabhive import ColabHive

client = ColabHive(api_key="hive_...", account_id="YOUR_ACCOUNT_ID")

result = client.endpoints.infer(
    endpoint_id="qwen-2.5-coder-7b-public",   # SDK resolves the name to a UUID
    input_data={"messages": [{"role": "user", "content":
        "Write a TypeScript function that filters an array of objects and returns sorted results, "
        "with type annotations."}]},
    max_tokens=600,
    temperature=0.2,
)

# Sync by default; a "queued" status means poll GET /tasks/{task_id}
print(result["result"] if result.get("status") != "queued"
      else client.endpoints.get_task(result["task_id"]))

Tips​

  • Use a low temperature (0.1–0.3) for deterministic code.
  • Provide related types/files as context, and ask for tests alongside the implementation.

Fine-tuning​

You can fine-tune a Qwen 2.5 Coder base on your own data with the llm-qlora-finetune / llm-qlora-peft training templates (transformers + PEFT + bitsandbytes backend). Point the base model at Qwen/Qwen2.5-Coder-7B-Instruct. See LLM Fine-Tuning and the Training API.

Next steps​

Authors: José Luis Minich, Maximiliano Lucius.