Hugging Face's logo

verifiers-for-code
/
phi3-deepseek-27k-cleanedplans-longtrain

PEFT
Safetensors
phi3
axolotl
Generated from Trainer
custom_code
Model card Files Community
phi3-deepseek-27k-cleanedplans-longtrain / README.md
sumuks's picture
sumuks
End of training
2c2ab17 verified
preview code
|
raw
history blame contribute delete
No virus
4.38 kB
metadata
base_model: microsoft/Phi-3-mini-4k-instruct
library_name: peft
license: mit
tags:
  - axolotl
  - generated_from_trainer
model-index:
  - name: phi3-deepseek-27k-cleanedplans-longtrain
    results: []

Built with Axolotl

See axolotl config

axolotl version: 0.4.1

# model and tokenizer
base_model: microsoft/Phi-3-mini-4k-instruct # change for model
trust_remote_code: true
sequence_len: 2048

strict: false

model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
bf16: auto
pad_to_sequence_len: true
save_safetensors: true


datasets:
  - path: verifiers-for-code/cleaned_deepseek_plans
    type: completion
    field: text
    train_on_split: train

val_set_size: 0.05

# lora
adapter: lora
lora_r: 2048
lora_alpha: 32
lora_dropout: 0.05
lora_target_linear: true
lora_modules_to_save:
  - embed_tokens
  - lm_head
use_rslora: true

# logging
wandb_project: valeris
wandb_name: phi3-deepseek-27k-cleanedplans-longtrain

output_dir: ./outputs/phi3-deepseek-27k-cleanedplans-longtrain

gradient_accumulation_steps: 2
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: true
micro_batch_size: 1
num_epochs: 3
eval_batch_size: 1
warmup_ratio: 0.05
learning_rate: 5e-6
lr_scheduler: cosine
optimizer: adamw_torch

hub_model_id: verifiers-for-code/phi3-deepseek-27k-cleanedplans-longtrain
push_to_hub: true
hub_always_push: true
evals_per_epoch: 8
saves_per_epoch: 4
logging_steps: 1
# eval_table_size: 10
# eval_max_new_tokens: 512

tokens: ["<thinking>", "</thinking>", "<plan>", "</plan>"]

special_tokens:
  pad_token: "<|endoftext|>"

phi3-deepseek-27k-cleanedplans-longtrain

This model is a fine-tuned version of microsoft/Phi-3-mini-4k-instruct on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.3618

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-06
  • train_batch_size: 1
  • eval_batch_size: 1
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 8
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 16
  • total_eval_batch_size: 8
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 242
  • num_epochs: 3

Training results

Training Loss Epoch Step Validation Loss
0.5446 0.0006 1 0.4877
0.4894 0.1255 203 0.4453
0.4329 0.2509 406 0.3950
0.4223 0.3764 609 0.3772
0.3909 0.5019 812 0.3705
0.3837 0.6273 1015 0.3676
0.3959 0.7528 1218 0.3658
0.3516 0.8782 1421 0.3642
0.3757 1.0037 1624 0.3632
0.3222 1.1292 1827 0.3627
0.3095 1.2546 2030 0.3624
0.3234 1.3801 2233 0.3621
0.3776 1.5056 2436 0.3620
0.3471 1.6310 2639 0.3618
0.343 1.7565 2842 0.3617
0.3898 1.8820 3045 0.3618
0.3207 2.0074 3248 0.3618
0.3486 2.1329 3451 0.3618
0.3362 2.2583 3654 0.3618
0.3444 2.3838 3857 0.3618
0.3717 2.5093 4060 0.3618
0.3482 2.6347 4263 0.3618
0.3393 2.7602 4466 0.3617
0.3121 2.8857 4669 0.3618

Framework versions

  • PEFT 0.11.1
  • Transformers 4.44.0.dev0
  • Pytorch 2.4.0
  • Datasets 2.19.1
  • Tokenizers 0.19.1