CoolFace
Modelpublic

chanceQZhang/zhihu-tech-zhichang-qwen-7b

sourceHugging Facegpl-3.0updated 9mo agoView on Hugging Face
0likes31downloads
Model Card

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/> <details><summary>See axolotl config</summary>

axolotl version: 0.13.0.dev0

yaml
# config_sft_zhihu.yml

base_model: Orion-zhen/Meissa-Qwen2.5-7B-Instruct
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

# 使用您上传的数据集
datasets:
  - path: chanceQZhang/zhihuhighvotes
    type: chat_template  # ChatML 格式使用 chat_template
    split: train
# 提速核心
sample_packing: true
pad_to_sequence_len: true



# LoRA 配置
adapter: lora
lora_r: 8
lora_alpha: 32
lora_dropout: 0.1
lora_target_modules:
  - q_proj
  - v_proj
  - k_proj
  - o_proj
  - gate_proj
  - up_proj
  - down_proj
  
# --- 核心优化：显存节省配置 ---
bf16: true                    # 30/40系列或A系列显卡必开，提升速度且省显存
fp16: false
gradient_checkpointing: true  # 必开！用计算时间换空间，大幅降低显存占用
flash_attention: true         # 必开！大幅降低长文本下的显存需求

# 训练配置
sequence_len: 2048
micro_batch_size: 6
gradient_accumulation_steps: 3
num_epochs: 2
learning_rate: 0.00005
# 减少中间开销
logging_steps: 10
eval_steps: 100
save_steps: 302

# 输出
output_dir: ./outputs/zhihu-tech-career-lora

</details><br>

outputs/zhihu-tech-career-lora

This model is a fine-tuned version of Orion-zhen/Meissa-Qwen2.5-7B-Instruct on the chanceQZhang/zhihuhighvotes dataset.

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • —learning_rate: 5e-05
  • —trainbatchsize: 6
  • —evalbatchsize: 6
  • —seed: 42
  • —gradientaccumulationsteps: 3
  • —totaltrainbatch_size: 18
  • —optimizer: Use OptimizerNames.ADAMWTORCHFUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
  • —lrschedulertype: cosine
  • —lrschedulerwarmup_steps: 16
  • —training_steps: 536

Training results

Framework versions

  • —PEFT 0.18.1
  • —Transformers 4.57.1
  • —Pytorch 2.8.0+cu128
  • —Datasets 4.4.2
  • —Tokenizers 0.22.2