How to use from the
Use from the
Transformers library
# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="YYYYYYibo/qwen3-4b-guru-scgate-k32-mb16-lr1e6-300-nvfp4-4k-step400")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)
# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("YYYYYYibo/qwen3-4b-guru-scgate-k32-mb16-lr1e6-300-nvfp4-4k-step400")
model = AutoModelForCausalLM.from_pretrained("YYYYYYibo/qwen3-4b-guru-scgate-k32-mb16-lr1e6-300-nvfp4-4k-step400", device_map="auto")
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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))
Quick Links

qwen3-4b-guru-scgate-k32-mb16-lr1e6-300-nvfp4-4k β€” step 400

Qwen3-4B-Base, guru-RL math, 4K context, NVFP4 rollouts. Method: Score Centering with ratio gate.

  • Run name: qwen3-4b-guru-scgate-k32-mb16-lr1e6-300-nvfp4-4k, checkpoint at training step 400.
  • Exported from verl FSDP actor checkpoint (full-precision master weights, no optimizer state). Rollouts during training used quantized vLLM (see run name), but these weights are the unquantized actor.
Downloads last month
297
Safetensors
Model size
4B params
Tensor type
BF16
Β·
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support