-
-
Notifications
You must be signed in to change notification settings - Fork 780
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Merge branch 'main' into 202312-better-devicemapping
- Loading branch information
Showing
2 changed files
with
81 additions
and
0 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,5 @@ | ||
# Overview | ||
|
||
This is an example of a Yi-34B-Chat configuration. It demonstrates that it is possible to finetune a 34B model on a GPU with 24GB of VRAM. | ||
|
||
Tested on an RTX 4090 with `python -m axolotl.cli.train examples/mistral/qlora.yml`, a single epoch of finetuning on the alpaca dataset using qlora runs in 47 mins, using 97% of available memory. |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,76 @@ | ||
base_model: 01-ai/Yi-34B-Chat | ||
model_type: LlamaForCausalLM | ||
tokenizer_type: LlamaTokenizer | ||
is_mistral_derived_model: false | ||
is_llama_derived_model: true | ||
load_in_8bit: false | ||
load_in_4bit: true | ||
strict: false | ||
sequence_len: 1024 | ||
bf16: true | ||
fp16: false | ||
tf32: false | ||
flash_attention: true | ||
special_tokens: | ||
bos_token: "<|startoftext|>" | ||
eos_token: "<|endoftext|>" | ||
unk_token: "<unk>" | ||
|
||
# Data | ||
datasets: | ||
- path: mhenrichsen/alpaca_2k_test | ||
type: alpaca | ||
warmup_steps: 10 | ||
|
||
# Iterations | ||
num_epochs: 1 | ||
|
||
# Evaluation | ||
val_set_size: 0.1 | ||
evals_per_epoch: 5 | ||
eval_table_size: | ||
eval_table_max_new_tokens: 128 | ||
eval_sample_packing: false | ||
eval_batch_size: 1 | ||
|
||
# LoRA | ||
output_dir: ./qlora-out | ||
adapter: qlora | ||
lora_model_dir: | ||
lora_r: 32 | ||
lora_alpha: 16 | ||
lora_dropout: 0.05 | ||
lora_target_linear: true | ||
lora_fan_in_fan_out: | ||
lora_target_modules: | ||
|
||
# Sampling | ||
sample_packing: false | ||
pad_to_sequence_len: false | ||
|
||
# Batching | ||
gradient_accumulation_steps: 4 | ||
micro_batch_size: 1 | ||
gradient_checkpointing: true | ||
|
||
# wandb | ||
wandb_project: | ||
|
||
# Optimizer | ||
optimizer: paged_adamw_8bit | ||
lr_scheduler: cosine | ||
learning_rate: 0.0002 | ||
|
||
# Misc | ||
train_on_inputs: false | ||
group_by_length: false | ||
early_stopping_patience: | ||
resume_from_checkpoint: | ||
local_rank: | ||
logging_steps: 1 | ||
xformers_attention: | ||
debug: | ||
deepspeed: | ||
weight_decay: 0 | ||
fsdp: | ||
fsdp_config: |