You can deploy this model and start earning money today!

kritokronos/
qwen35-4b-claude46-opus-reasoning-merged

Name: qwen35-4b-claude46-opus-reasoning-merged
Availability: InStock

Views

No views yet

Qwen3.5-4B Claude-4.6-Opus-Reasoning Merged

This repository contains a merged reasoning-oriented language model based on Qwen3.5-4B.

The model was fine-tuned using supervised fine-tuning / distillation-style data to improve step-by-step reasoning, instruction following, and chat-style response quality.

This model is not affiliated with Anthropic, Claude, or OpenAI. The model name only describes the training target/style used in the distillation process.

Model Details

Base model: Qwen/Qwen3.5-4B
Model type: Causal Language Model
Training method: Supervised Fine-Tuning / Distillation
Checkpoint type: Merged model
Primary language: English
Library: Hugging Face Transformers
Intended use: reasoning, instruction following, educational QA, coding/math-style reasoning experiments

Training Details

The model was fine-tuned from the Qwen3.5-4B base model using reasoning-style instruction data.

Training setup:

Fine-tuning method: LoRA / SFT, then merged into the base model
Framework: Transformers, PEFT, TRL
Precision: bf16 / fp16
Sequence length: 512 tokens
Objective: next-token prediction on formatted chat/instruction data
Dataset: nohurry/Opus-4.6-Reasoning-3000x-filtered

Prompt Format

This model follows the Qwen chat template format:

text

<|im_start|>user
Your question here
<|im_end|>
<|im_start|>assistant

Recommended usage is through tokenizer.apply_chat_template.

Usage

python

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "kritokronos/qwen35-4b-claude46-opus-reasoning-merged"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True,
)

messages = [
    {
        "role": "user",
        "content": "If a bus travels 60 km in 1.5 hours, what is its average speed?"
    }
]

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)

inputs = tokenizer(text, return_tensors="pt").to(model.device)

terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|im_end|>")
]

outputs = model.generate(
    **inputs,
    max_new_tokens=512,
    temperature=0.6,
    top_p=0.9,
    repetition_penalty=1.1,
    eos_token_id=terminators,
)

response = tokenizer.decode(
    outputs[0][inputs["input_ids"].shape[-1]:],
    skip_special_tokens=True,
)

print(response)