This is a
LoRA adapter
, not a standalone model. Load it on the pinned
Qwen/Qwen3-1.7B-Base
revision
ea980cb0a6c2ae4b936e82123acc929f1cec04c1
. It belongs to the
Qwen3 Code Post-Training Lab
, a hands-on project comparing code post-training methods under one evaluation contract.
这是用于 Python 函数补全的 LoRA SFT 适配器。输入应是原始代码题干，不是聊天消息；只下载本仓库的适配器文件无法独立推理，还需下载锁定版本的 Base 权重。
Training
Same 26,805 training and 1,386 validation examples as the project's Full SFT model, filtered from a fixed shard of
NVIDIA OpenCodeInstruct
(CC BY 4.0). Exact source revision and attribution:
data lock
.
Single RTX 4090; 1 epoch, 1,676 optimizer steps, effective batch size 16, bf16, max sequence length 1,024.
LoRA rank 16, alpha 32, dropout 0.05, attention and MLP linear layers; learning rate
1e-4
. Loss is computed only on the code completion.
Best validation loss:
0.14723
. The adapter was selected by validation loss, not benchmark score.
Evaluation
Benchmark
Base
Full SFT
This LoRA SFT adapter
HumanEval+ v0.1.10
31/164 (18.9%)
67/164 (40.9%)
82/164 (50.0%)
MBPP+ v0.2.0
214/378 (56.6%)
229/378 (60.6%)
237/378 (62.7%)
Strict pass@1 requires both original and Plus tests to pass. The same frozen tasks, raw prompt, greedy decoding, 512-token cap and restricted Docker judge were used for all three models. The different SFT learning rates and optimizers mean this result should not be attributed solely to the LoRA update mechanism.
Per-task results and evaluation details
are public.
Load and generate
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base = AutoModelForCausalLM.from_pretrained(
"Qwen/Qwen3-1.7B-Base",
revision="ea980cb0a6c2ae4b936e82123acc929f1cec04c1",
dtype="auto",
device_map="auto",
)
adapter = "Eternity5551/Qwen3-1.7B-Python-Code-LoRA-SFT"
model = PeftModel.from_pretrained(base, adapter).eval()
tokenizer = AutoTokenizer.from_pretrained(adapter)
prompt = '"""\nWrite a Python function double_even(nums) that doubles only even integers.\n"""\n\n'
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
output = model.generate(**inputs, do_sample=False, max_new_tokens=512,
pad_token_id=tokenizer.eos_token_id)
print(tokenizer.decode(output[0, inputs.input_ids.shape[1]:], skip_special_tokens=True))
Limitations and credits
These are public Python function benchmarks and cannot guarantee performance on unseen code. Filtering by exact word spans cannot rule out all semantic overlap. Base model: Qwen team, Apache 2.0. Training data: NVIDIA OpenCodeInstruct, CC BY 4.0. The adapter weights were verified byte-for-byte against the training output before upload.
Runs of Eternity5551 Qwen3-1.7B-Python-Code-LoRA-SFT on huggingface.co
21
Total runs
0
24-hour runs
0
3-day runs
2
7-day runs
21
30-day runs
More Information About Qwen3-1.7B-Python-Code-LoRA-SFT huggingface.co Model
More Qwen3-1.7B-Python-Code-LoRA-SFT license Visit here:
Qwen3-1.7B-Python-Code-LoRA-SFT huggingface.co is an AI model on huggingface.co that provides Qwen3-1.7B-Python-Code-LoRA-SFT's model effect (), which can be used instantly with this Eternity5551 Qwen3-1.7B-Python-Code-LoRA-SFT model. huggingface.co supports a free trial of the Qwen3-1.7B-Python-Code-LoRA-SFT model, and also provides paid use of the Qwen3-1.7B-Python-Code-LoRA-SFT. Support call Qwen3-1.7B-Python-Code-LoRA-SFT model through api, including Node.js, Python, http.
Qwen3-1.7B-Python-Code-LoRA-SFT huggingface.co is an online trial and call api platform, which integrates Qwen3-1.7B-Python-Code-LoRA-SFT's modeling effects, including api services, and provides a free online trial of Qwen3-1.7B-Python-Code-LoRA-SFT, you can try Qwen3-1.7B-Python-Code-LoRA-SFT online for free by clicking the link below.
Eternity5551 Qwen3-1.7B-Python-Code-LoRA-SFT online free url in huggingface.co:
Qwen3-1.7B-Python-Code-LoRA-SFT is an open source model from GitHub that offers a free installation service, and any user can find Qwen3-1.7B-Python-Code-LoRA-SFT on GitHub to install. At the same time, huggingface.co provides the effect of Qwen3-1.7B-Python-Code-LoRA-SFT install, users can directly use Qwen3-1.7B-Python-Code-LoRA-SFT installed effect in huggingface.co for debugging and trial. It also supports api for free installation.
Qwen3-1.7B-Python-Code-LoRA-SFT install url in huggingface.co:
