ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Revert

Browse source
This commit is contained in:
Daniel Han 2025-03-13 05:13:17 -07:00
commit 9f89844a81
3 changed files with 49 additions and 41 deletions
Download patch file Download diff file Expand all files Collapse all files

38
unsloth/models/loader.py
View file

@ -609,30 +609,30 @@ class FastModel(FastBaseModel):
patch_loss_functions(torch_compile = False)
model_types = unsloth_compile_transformers(
model_name = model_name,
sdpa_dynamic_mask = False,
sdpa_bool_masks = False,
sdpa_gqa_replace = False,
sdpa_dynamic_compile = False,
compile_attention = False,
disable_causal_masks = False,
compile_torch_modules = False,
compile_custom_modules = False,
compile_function_calls = False,
fuse_lm_head = False,
gradient_checkpointing = False,
manual_replacements = False,
fast_lora_forwards = False,
sdpa_dynamic_mask = True,
sdpa_bool_masks = True,
sdpa_gqa_replace = True,
sdpa_dynamic_compile = True,
compile_attention = True,
disable_causal_masks = True,
compile_torch_modules = True,
compile_custom_modules = True,
compile_function_calls = True,
fuse_lm_head = True,
gradient_checkpointing = True,
manual_replacements = True,
fast_lora_forwards = True,
fast_residual_stream = False,
accurate_accumulation = False,
epilogue_fusion = False,
accurate_accumulation = True,
epilogue_fusion = True,
max_autotune = False,
shape_padding = False,
shape_padding = True,
cudagraphs = False,
debug = False,
fullgraph = False,
fullgraph = fullgraph,
import_from_cache = False,
disable = False,
return_logits = False,
return_logits = return_logits,
)
pass
@ -668,7 +668,7 @@ class FastModel(FastBaseModel):
use_gradient_checkpointing = use_gradient_checkpointing,
*args, **kwargs,
)
if resize_model_vocab is not None:
model.resize_token_embeddings(resize_model_vocab)
pass

24
unsloth/models/mapper.py
View file

@ -638,37 +638,45 @@ __INT_TO_FLOAT_MAPPER = \
"Qwen/QwQ-32B",
"unsloth/QwQ-32B-bnb-4bit",
),
"unsloth/gemma-3-1b-it-bnb-4bit" : (
"unsloth/gemma-3-1b-it-unsloth-bnb-4bit" : (
"unsloth/gemma-3-1b-it",
"google/gemma-3-1b-it",
"unsloth/gemma-3-1b-it-bnb-4bit",
),
"unsloth/gemma-3-4b-it-bnb-4bit" : (
"unsloth/gemma-3-4b-it-unsloth-bnb-4bit" : (
"unsloth/gemma-3-4b-it",
"google/gemma-3-4b-it",
"unsloth/gemma-3-4b-it-bnb-4bit",
),
"unsloth/gemma-3-12b-it-bnb-4bit" : (
"unsloth/gemma-3-12b-it-unsloth-bnb-4bit" : (
"unsloth/gemma-3-12b-it",
"google/gemma-3-12b-it",
"unsloth/gemma-3-12b-it-bnb-4bit",
),
"unsloth/gemma-3-27b-it-bnb-4bit" : (
"unsloth/gemma-3-27b-it-unsloth-bnb-4bit" : (
"unsloth/gemma-3-27b-it",
"google/gemma-3-27b-it",
"unsloth/gemma-3-27b-it-bnb-4bit",
),
"unsloth/gemma-3-1b-pt-bnb-4bit" : (
"unsloth/gemma-3-1b-pt-unsloth-bnb-4bit" : (
"unsloth/gemma-3-1b-pt",
"google/gemma-3-1b-pt",
"unsloth/gemma-3-1b-pt-bnb-4bit",
),
"unsloth/gemma-3-4b-pt-bnb-4bit" : (
"unsloth/gemma-3-4b-pt-unsloth-bnb-4bit" : (
"unsloth/gemma-3-4b-pt",
"google/gemma-3-4b-pt",
"unsloth/gemma-3-4b-pt-bnb-4bit",
),
"unsloth/gemma-3-12b-pt-bnb-4bit" : (
"unsloth/gemma-3-12b-pt-unsloth-bnb-4bit" : (
"unsloth/gemma-3-12b-pt",
"google/gemma-3-12b-pt",
"unsloth/gemma-3-12b-pt-bnb-4bit",
),
"unsloth/gemma-3-27b-pt-bnb-4bit" : (
"unsloth/gemma-3-27b-pt-unsloth-bnb-4bit" : (
"unsloth/gemma-3-27b-pt",
"google/gemma-3-27b-pt",
"unsloth/gemma-3-27b-pt-bnb-4bit",
),
}

28
unsloth/models/vision.py
View file

@ -245,22 +245,22 @@ class FastBaseModel:
auto_processor = AutoProcessor if auto_model is AutoModelForVision2Seq else AutoTokenizer
tokenizer = auto_processor.from_pretrained(
tokenizer_name,
padding_side = "left",
padding_side = "right",
token = token,
)
# Add padding side as well
if hasattr(tokenizer, "tokenizer"):
tokenizer.tokenizer.padding_side = "left"
tokenizer.tokenizer.padding_side = "right"
# model, tokenizer = patch_tokenizer(model, tokenizer)
# model = post_patch_loss_function(model)
model, tokenizer = patch_tokenizer(model, tokenizer)
model = post_patch_loss_function(model)
# Fix other stuff like BnB compute data types
# model, tokenizer = patch_model_and_tokenizer(
# model,
# tokenizer,
# downcast_rope = False,
# fix_embeddings = False,
# )
model, tokenizer = patch_model_and_tokenizer(
model,
tokenizer,
downcast_rope = False,
fix_embeddings = False,
)
# Log Unsloth version for future fastpaths for inference
if hasattr(model, "config"):
@ -297,10 +297,10 @@ class FastBaseModel:
model.generate = types.MethodType(unsloth_base_fast_generate, model)
# Post patches
# model = FastBaseModel.post_patch_model(
# model,
# use_gradient_checkpointing = use_gradient_checkpointing,
# )
model = FastBaseModel.post_patch_model(
model,
use_gradient_checkpointing = use_gradient_checkpointing,
)
# Clear deleted GPU items
for _ in range(3):
gc.collect()