IQuest-Coder-V1
State-of-the-art code LLM family built on the innovative Code-Flow training paradigm. Achieves 76.2% on SWE-Bench Verified with native 128K context.
Key Highlights
State-of-the-Art
76.2% SWE-Bench Verified, 81.1% LiveCodeBench v6, 49.9% BigCodeBench.
Code-Flow Training
Learns from repository evolution, commits, and dynamic code transformations.
Dual Specialization
Thinking: Reasoning RL. Instruct: Fast coding assistance.
128K Context
Native long context without additional scaling tricks.
Model Family
| Model | Parameters | Context | Variants |
|---|---|---|---|
| IQuest-Coder-V1-7B | 7B | 128K | Instruct, Thinking |
| IQuest-Coder-V1-14B | 14B | 128K | Instruct, Thinking |
| IQuest-Coder-V1-40B | 40B | 128K | Instruct, Thinking, Loop |
💡 Sampling: Temperature=0.6, TopP=0.85, TopK=20 for Instruct models.
Quickstart
Requires transformers>=4.52.4.
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "IQuest/IQuest-Coder-V1-40B-Instruct"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto"
)
prompt = "Write a Python function to calculate Fibonacci."
messages = [{"role": "user", "content": prompt}]
text = tokenizer.apply_chat_template(
messages, tokenize=False, add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(**model_inputs, max_new_tokens=8192)
response = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
print(response)
Deployment (vLLM)
For production, use vLLM for high-throughput serving.
Instruct Model
vllm serve IQuestLab/IQuest-Coder-V1-40B-Instruct --tensor-parallel-size 8
Thinking Model (with reasoning)
vllm serve IQuestLab/IQuest-Coder-V1-40B-Thinking --reasoning-parser qwen3 --tensor-parallel-size 8