mirror/OpenLLM
1
0
Fork 0
mirror of https://github.com/bentoml/OpenLLM.git synced 2026-04-23 08:28:24 -04:00
Code Issues Packages Projects Releases Wiki Activity

chore: ignore new lines split [skip ci]

Browse Source 
Signed-off-by: aarnphm-ec2-dev <29749331+aarnphm@users.noreply.github.com>
This commit is contained in:
aarnphm-ec2-dev
2023-09-01 17:00:49 +00:00
parent 608de0b667
commit 7d893e6cd2
70 changed files with 575 additions and 950 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
openllm-python/src/openllm/playground/falcon_tuned.py
View File

@@ -56,18 +56,13 @@ if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
else:
model_args, training_args = t.cast(t.Tuple[ModelArguments, TrainingArguments], parser.parse_args_into_dataclasses())
model, tokenizer = openllm.AutoLLM.for_model("falcon",
model_id=model_args.model_id,
quantize="int4",
bnb_4bit_quant_type="nf4",
bnb_4bit_compute_dtype=torch.float16,
ensure_available=True).prepare_for_training(
adapter_type="lora",
lora_alpha=16,
lora_dropout=0.1,
r=16,
bias="none",
target_modules=["query_key_value", "dense", "dense_h_to_4h", "dense_4h_to_h"])
model, tokenizer = openllm.AutoLLM.for_model("falcon", model_id=model_args.model_id, quantize="int4", bnb_4bit_quant_type="nf4", bnb_4bit_compute_dtype=torch.float16,
ensure_available=True).prepare_for_training(adapter_type="lora",
lora_alpha=16,
lora_dropout=0.1,
r=16,
bias="none",
target_modules=["query_key_value", "dense", "dense_h_to_4h", "dense_4h_to_h"])
model.config.use_cache = False
tokenizer.pad_token = tokenizer.eos_token

11
openllm-python/src/openllm/playground/llama2_qlora.py
View File

@@ -98,8 +98,7 @@ def prepare_datasets(tokenizer, dataset_name=DATASET_NAME):
print("Sample from dolly-v2 ds:", dataset[randint(0, len(dataset))]["text"])
# tokenize and chunk dataset
lm_dataset = dataset.map(lambda sample: tokenizer(sample["text"]), batched=True,
remove_columns=list(dataset.features)).map(partial(chunk, chunk_length=2048), batched=True)
lm_dataset = dataset.map(lambda sample: tokenizer(sample["text"]), batched=True, remove_columns=list(dataset.features)).map(partial(chunk, chunk_length=2048), batched=True)
# Print total number of samples
print(f"Total number of samples: {len(lm_dataset)}")
@@ -180,15 +179,11 @@ def train_loop(model_args: ModelArguments, training_args: TrainingArguments):
transformers.set_seed(model_args.seed)
model, tokenizer = prepare_for_int4_training(model_args.model_id,
gradient_checkpointing=training_args.gradient_checkpointing,
bf16=training_args.bf16,
)
model, tokenizer = prepare_for_int4_training(model_args.model_id, gradient_checkpointing=training_args.gradient_checkpointing, bf16=training_args.bf16,)
datasets = prepare_datasets(tokenizer)
trainer = transformers.Trainer(model=model,
args=dataclasses.replace(transformers.TrainingArguments(training_args.output_dir),
**dataclasses.asdict(training_args)),
args=dataclasses.replace(transformers.TrainingArguments(training_args.output_dir), **dataclasses.asdict(training_args)),
train_dataset=datasets,
data_collator=transformers.default_data_collator,
)

13
openllm-python/src/openllm/playground/opt_tuned.py
View File

@@ -56,13 +56,12 @@ if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
else:
model_args, training_args = t.cast(t.Tuple[ModelArguments, TrainingArguments], parser.parse_args_into_dataclasses())
model, tokenizer = openllm.AutoLLM.for_model("opt", model_id=model_args.model_id, quantize="int8",
ensure_available=True).prepare_for_training(adapter_type="lora",
r=16,
lora_alpha=32,
target_modules=["q_proj", "v_proj"],
lora_dropout=0.05,
bias="none")
model, tokenizer = openllm.AutoLLM.for_model("opt", model_id=model_args.model_id, quantize="int8", ensure_available=True).prepare_for_training(adapter_type="lora",
r=16,
lora_alpha=32,
target_modules=["q_proj", "v_proj"],
lora_dropout=0.05,
bias="none")
# ft on english_quotes
data = load_dataset("Abirate/english_quotes")