Downloads · 30 days
5
8% of all-time downloads
diegogs1451/qwen2.5-7B-Instruct-dUO-finetuned-20260703-3epochs
qwen2.5-7B-Instruct-dUO-finetuned-20260703-3epochs is a machine learning model from diegogs1451. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for peft.
This repository contains a LoRA adapter fine-tuned from Qwen/Qwen2.5-7B-Instruct for causal language modeling. The saved artifacts indicate a 3-epoch run with a small-rank adapter and evaluation tracked by mean token…
Downloads · 30 days
5
8% of all-time downloads
All-time downloads
62
Public
Repo size
739 MB
Likes
0
Public
Click a slice to open those files.
.pt485 MB · 56%
From the Hugging Face model README
This repository contains a LoRA adapter fine-tuned from Qwen/Qwen2.5-7B-Instruct for causal language modeling. The saved artifacts indicate a 3-epoch run with a small-rank adapter and evaluation tracked by mean token accuracy and loss.
This is a PEFT LoRA adapter on top of Qwen/Qwen2.5-7B-Instruct. The adapter targets the attention and MLP projection layers of the base model and is intended to be loaded together with the base checkpoint at inference time.
Use this adapter with the Qwen2.5-7B-Instruct base model for text generation, instruction following, and other causal language modeling tasks that benefit from the fine-tuning distribution used during training.
Suitable for applications that want the behavior captured by the fine-tuning run, such as domain-specific chat, drafting, or response shaping. The adapter should be merged or loaded at inference time with the matching base model.
Do not use this adapter as a standalone model. It is not intended for safety-critical decisions, medical or legal advice, or any setting that requires guaranteed factual accuracy.
The adapter inherits the base model’s limitations and may amplify them on the fine-tuning distribution. It can still hallucinate, produce unsafe or biased outputs, and perform poorly outside the domain covered by training data. The repository does not include an explicit dataset card, so the exact training corpus and filtering policy are not documented here.
Review outputs before use in production, add application-level safety checks, and validate behavior on your own target data before deployment.
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base_model_id = "Qwen/Qwen2.5-7B-Instruct"
adapter_path = "."
tokenizer = AutoTokenizer.from_pretrained(base_model_id, use_fast=True)
model = AutoModelForCausalLM.from_pretrained(base_model_id, device_map="auto")
model = PeftModel.from_pretrained(model, adapter_path)
The exact dataset is not recorded in the saved artifacts. Based on the trainer logs, the run processed at least 279,929 training tokens by the end of training.
Not documented in the repository.
The final checkpoint was reached at global step 150. The last recorded evaluation metrics were eval loss 0.1313 and eval mean token accuracy 0.9631.
Not documented in the saved artifacts.
Not documented in the saved artifacts.
Mean token accuracy and evaluation loss.
At the final recorded checkpoint, evaluation loss was 0.13125666975975037 and evaluation mean token accuracy was 0.96309861770043. Training loss at the final logged step was 0.1333 with mean token accuracy 0.9644.
The adapter converged steadily over 3 epochs and produced consistent token-level accuracy above 0.96 on the recorded evaluation runs.
No interpretability artifacts are included.
Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).
LoRA adapter for a causal language model. The adapter is configured for Qwen/Qwen2.5-7B-Instruct and updates the attention and MLP projection layers during fine-tuning.
Not documented in the saved artifacts.
Not documented.
Transformers-based training with PEFT 0.15.2.
BibTeX:
Not provided.
APA:
Not provided.
LoRA: Low-Rank Adaptation, a parameter-efficient fine-tuning method.
PEFT: Parameter-Efficient Fine-Tuning.
This card was reconstructed from the saved adapter, tokenizer, and trainer state files in the repository.
Not specified.
Not specified.