Instructions to use RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random")
model = AutoModelForCausalLM.from_pretrained("RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random

SGLang

How to use RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random with Docker Model Runner:
```
docker model run hf.co/RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random
```

qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random

A 4B math-distilled model. Student fine-tuned from RockToken/qwen3_30b_a3b_to_4b_onpolicy_5k_src20k-25k_freeze_random via on-policy reverse-KL distillation against a Qwen3-30B-A3B teacher, using the token_freeze_kd algorithm to mask a 98-token "freeze list" out of the KD loss.

This is the control arm ("random" token set) of an A/B experiment on the effect of selectively freezing certain tokens during on-policy KD.

Training


Student (init)	`RockToken/qwen3_30b_a3b_to_4b_onpolicy_5k_src20k-25k_freeze_random`
Teacher	`Qwen/Qwen3-30B-A3B-Instruct-2507`
Data	OpenThoughts3 math prompts — 10k cumulative (5k src20k-25k → continued 5k src25k-30k)
Source file	`openthoughts_prompt_math_5k_src25k-30k.jsonl (continuation)`
KD algorithm	`token_freeze_kd` (see KDFlow)
KD loss	reverse-KL, kd_temperature=1.0, kd_ratio=1.0
Freeze list	98 token IDs from `random.json` (control set)
Freeze weight	0.0 (loss on these tokens is fully zeroed)
Backend	FSDP2, bf16, gradient checkpointing
Topology	4× H100 (1 node), teacher TP=4, rollout TP=2
Optimizer	AdamW, lr=2e-6 (5% warmup, cosine→1e-8), max_norm=1.0
Batch	train_batch_size=4, micro=1, n_samples_per_prompt=4
Steps	1 epoch, 2500 steps
Rollout	response cap 1024 tokens, temperature=1.0, top_p=1.0
Chat template	applied; `enable_thinking=False` (Instruct-2507 is non-thinking)

A/B siblings

5k variants (this slice only):
- RockToken/qwen3_30b_a3b_to_4b_onpolicy_5k_src20k-25k_freeze_random — control
- RockToken/qwen3_30b_a3b_to_4b_onpolicy_5k_src20k-25k_freeze_rock — treatment
10k variants (5k + continued 5k on src25k-30k):
- RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random — control
- RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_rock — treatment

Use

from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random")
model = AutoModelForCausalLM.from_pretrained("RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random", torch_dtype="bfloat16")

Downloads last month: 15

Safetensors

Model size

4B params

Tensor type

BF16

Model tree for RockToken/qwen3_30b_a3b_to_4b_onpolicy_10k_src20k-30k_freeze_random

Base model

RockToken/qwen3_30b_a3b_to_4b_offpolicy_20k

Finetuned

RockToken/qwen3_30b_a3b_to_4b_onpolicy_5k_src20k-25k_freeze_random

Finetuned

(1)

this model