llm_surgery

Autometa's group workspace

Group: Mistral-7B-v0.1

1-2

of 2

Tags

align-dpo

Author

capecape

State

Finished

Start time

March 10th, 2024 11:44:30 PM

Runtime

32m 1s

Tracked hours

30m 37s

Run path

llm_surgery/gemma-zephyr/ds5rs06h

Linux-5.15.0-1047-oracle-x86_64-with-glibc2.35

Python version

3.11.7

Git repository

git clone git@github.com:tcapelle/alignment-handbook.git

Git state

git checkout -b "Mistral-7B-v0.1_dpo" 4de29b78c28e96ecb5a1c8144d65e398666f8f12

Command

/workspace/scripts/run_dpo.py recipes/zephyr-7b-beta/dpo/config_full.yaml --model_name_or_path=llm_surgery/gemma-zephyr/Mistral-7B-v0.1:v7

System Hardware

CPU count	128
Logical CPU count	255
GPU count	8
GPU type	NVIDIA A100-SXM4-80GB

W&B CLI Version

0.16.3

Group

Mistral-7B-v0.1

Job Type

train-dpo

Config parameters are your model's inputs. Learn more

▶
Config parameters:{} 192 keys
- _name_or_path:
  "/workspace/artifacts/Mistral-7B-v0.1:v7"
- ▶
  accelerator_config:{} 4 keys
  - dispatch_batches:
    null
  - even_batches:
    true
  - split_batches:
    false
  - use_seedable_sampler:
    true
- adafactor:
  false
- adam_beta1:
  0.9
- adam_beta2:
  0.999
- adam_epsilon:
  0.00000001
- add_cross_attention:
  false
- ▶
  architectures:[] 1 item
  - 0:
    "MistralForCausalLM"
- attention_dropout:
  0
- auto_find_batch_size:
  false
- bad_words_ids:
  null
- begin_suppress_tokens:
  null
- beta:
  0.05
- bf16:
  true
- bf16_full_eval:
  false
- bos_token_id:
  1
- chunk_size_feed_forward:
  0
- cross_attention_hidden_size:
  null
- data_seed:
  null
- dataloader_drop_last:
  false
- dataloader_num_workers:
  0
- dataloader_persistent_workers:
  false
- dataloader_pin_memory:
  true
- dataloader_prefetch_factor:
  null
- ddp_backend:
  null
- ddp_broadcast_buffers:
  null
- ddp_bucket_cap_mb:
  null
- ddp_find_unused_parameters:
  null
- ddp_timeout:
  1,800
- debug:[] 0 items
- decoder_start_token_id:
  null
- deepspeed:
  null
- disable_tqdm:
  false
- dispatch_batches:
  null
- diversity_penalty:
  0
- do_eval:
  true
- do_predict:
  false
- do_sample:
  false
- do_train:
  false
- early_stopping:
  false
- encoder_no_repeat_ngram_size:
  0
- eos_token_id:
  2
- eval_accumulation_steps:
  null
- eval_delay:
  0
- eval_steps:
  100
- evaluation_strategy:
  "steps"
- vocab_size:
  32,032
- warmup_ratio:
  0.1
- warmup_steps:
  0
- weight_decay:
  0

Summary metrics are your model's outputs. Learn more

▶
Summary metrics:{} 30 keys
- eval/logits/chosen:
  -1.5737099647521973
- eval/logits/rejected:
  -1.4398664236068726
- eval/logps/chosen:
  -292.6082458496094
- eval/logps/rejected:
  -318.71435546875
- eval/loss:
  0.46746960282325745
- eval/rewards/accuracies:
  0.75
- eval/rewards/chosen:
  -0.7094362378120422
- eval/rewards/margins:
  1.2131987810134888
- eval/rewards/rejected:
  -1.922635197639465
- eval/runtime:
  37.0405
- eval/samples_per_second:
  20.248
- eval/steps_per_second:
  0.648
- train/epoch:
  1.97
- train/global_step:
  104
- train/grad_norm:
  28.72763529396233
- train/learning_rate:
  0
- train/logits/chosen:
  -1.996106505393982
- train/logits/rejected:
  -1.91734778881073
- train/logps/chosen:
  -407.35638427734375
- train/logps/rejected:
  -470.3741760253906
- train/loss:
  0.2314
- train/rewards/accuracies:
  0.9375
- train/rewards/chosen:
  -0.12680268287658691
- train/rewards/margins:
  2.807525873184204
- train/rewards/rejected:
  -2.934328556060791
- train/total_flos:
  0
- train/train_loss:
  0.3958657637525063
- train/train_runtime:
  1,695.8776
- train/train_samples_per_second:
  7.96
- train/train_steps_per_second:
  0.061

This run consumed these artifacts as inputs. Learn more

This run produced these artifacts as outputs. Total: 2. Learn more

model

Mistral-7B-v0.1:v9

wandb-history

run-ds5rs06h-history:v0