metadata
license: apache-2.0
base_model: Qwen/Qwen3.6-35B-A3B
tags:
- lora
- peft
- qwen
- qwen3
- function-calling
- tool-use
- xlam
datasets:
- minpeter/xlam-function-calling-60k-parsed
library_name: peft
Qwen3.6-35B-A3B xLAM Function-Calling LoRA
A LoRA adapter fine-tuned on minpeter/xlam-function-calling-60k-parsed from the Qwen/Qwen3.6-35B-A3B base model.
Training configuration
- Method: LoRA (bf16)
- LoRA rank (r): 64
- LoRA alpha: 128
- Dataset: minpeter/xlam-function-calling-60k-parsed
- Infrastructure: NVIDIA Run:AI on Amazon SageMaker HyperPod (EKS mode)
Usage with transformers + peft
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch
base = AutoModelForCausalLM.from_pretrained(
"Qwen/Qwen3.6-35B-A3B",
torch_dtype=torch.bfloat16,
device_map="auto",
)
model = PeftModel.from_pretrained(base, "ying2022/qwen3-6-35b-xlam-tools-lora")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3.6-35B-A3B")
messages = [{"role": "user", "content": "<your prompt here>"}]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to(model.device)
outputs = model.generate(inputs, max_new_tokens=500, temperature=0.3)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Serving with vLLM
vllm serve Qwen/Qwen3.6-35B-A3B \
--enable-lora \
--lora-modules adapter=ying2022/qwen3-6-35b-xlam-tools-lora \
--max-loras 1 --max-lora-rank 64 \
--tensor-parallel-size 2 \
--gpu-memory-utilization 0.90 \
--max-model-len 4096