Skip to content

[not4land] temp change to convert torchao checkpoint to gguf #504

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[not4land] temp change to convert torchao checkpoint to gguf #504

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
132 changes: 88 additions & 44 deletions auto_round/export/export_to_gguf/convert.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -755,6 +755,8 @@ def get_vocab_base_pre(self, tokenizer) -> str:
if chkhsh == "877081d19cf6996e2c4ff0e1236341e9b7bde288f5311a56a937f0afbbb3aeb5":
# ref: https://huggingface.co/deepseek-ai/DeepSeek-V3
res = "deepseek-v3"
if chkhsh == "ccc2ef013c104be7bae2965776d611e1d7a8a2a9c547dd93a682c9a9fc80352e":
res = "gpt2"

if res is None:
logger.warning("\n")
Expand Down Expand Up @@ -1073,7 +1075,10 @@ def prepare_tensors(self):
break

for new_name, data_torch in (self.modify_tensors(data_torch, name, bid)):
data = data_torch.squeeze().cpu().numpy()
if type(data_torch) in [torch.Tensor, torch.nn.Parameter]:
data = data_torch.squeeze().cpu().numpy()
else:
data = data_torch

# if data ends up empty, it means data_torch was a scalar tensor -> restore
if len(data.shape) == 0:
Expand Down Expand Up @@ -1154,14 +1159,16 @@ def _quant_data(data, data_qtype):
bs = module.scale.shape[0]
for attr in ["scale", "zp", "w_d_scale", "w_d_wmin_m", "w_wmin_m"]:
if hasattr(module, attr) and getattr(module, attr) is not None:
print("quant data")
breakpoint()
attr_tensor = getattr(module, attr)
ori_shape = attr_tensor.shape
attr_tensor = self.modify_tensors(attr_tensor.reshape(bs, -1), name, bid)[0][1]
attr_tensor = attr_tensor.reshape(ori_shape)
setattr(module, attr, attr_tensor)

scale = module.scale

if isinstance(scale, torch.Tensor):
scale = scale.numpy()
zp = module.zp if hasattr(module, "zp") else None
Expand All @@ -1187,6 +1194,40 @@ def _quant_data(data, data_qtype):
data_qtype = gguf.GGMLQuantizationType.F32
return data, data_qtype

def _torchao_quant_data(data):
data_qtype = gguf.GGMLQuantizationType.Q4_K
from auto_round.utils import get_module
suffix = '.weight'
if suffix in name:
layer_name = name[:-len(suffix)]
module = get_module(self.model, layer_name)
d_scale = data.qparams_dict["d_scale"].cpu().numpy()
d_wmin_m = data.qparams_dict["d_wmin_m"].cpu().numpy()
scale = data.qparams_dict["scale"].cpu().numpy()
wmin_m = data.qparams_dict["wmin_m"].cpu().numpy()
# attr_to_tensor = {
# "scale": quantized_block_scale,
# "w_wmin_m": quantized_block_min,
# "w_d_scale": super_block_scale_scale,
# "w_d_wmin_m": super_block_min_scale,
# }
# for attr_name, attr_tensor in attr_to_tensor.items():
# ori_shape = attr_tensor.shape
# attr_tensor = self.modify_tensors(attr_tensor.reshape(bs, -1), name, bid)[0][1]
# attr_tensor = attr_tensor.reshape(ori_shape)
# setattr(module, attr_name, attr_tensor)
# int_data = data.int_data.cpu().numpy()
float_data = data.qparams_dict["orig_tensor"].cpu().numpy()
data = ggml_quant(
float_data,
data_qtype.name.lower(),
scale,
None,
wmin_m=wmin_m,
d_scale=d_scale,
d_wmin_m=d_wmin_m)
return data, data_qtype

# for MOE model
if len(data.shape) == 3:
new_data = []
Expand All @@ -1201,7 +1242,10 @@ def _quant_data(data, data_qtype):
data = np.array(new_data)
del new_data
else:
data, data_qtype = _quant_data(data, data_qtype)
if type(data) == np.ndarray:
data, data_qtype = _quant_data(data, data_qtype)
else:
data, data_qtype = _torchao_quant_data(data)

shape = gguf.quant_shape_from_byte_shape(
data.shape, data_qtype) if data.dtype == np.uint8 else data.shape
Expand Down Expand Up @@ -2858,61 +2902,61 @@ def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
torch.tensor(short_factors, dtype=torch.float32))


@Model.register("PhiMoEForCausalLM")
class PhiMoeModel(Phi3MiniModel):
model_arch = gguf.MODEL_ARCH.PHIMOE
# @Model.register("PhiMoEForCausalLM")
# class PhiMoeModel(Phi3MiniModel):
# model_arch = gguf.MODEL_ARCH.PHIMOE

_experts: list[dict[str, Tensor]] | None = None
# _experts: list[dict[str, Tensor]] | None = None

def set_gguf_parameters(self):
super().set_gguf_parameters()
self.gguf_writer.add_expert_used_count(self.hparams["num_experts_per_tok"])
self.gguf_writer.add_expert_count(self.hparams["num_local_experts"])
# def set_gguf_parameters(self):
# super().set_gguf_parameters()
# self.gguf_writer.add_expert_used_count(self.hparams["num_experts_per_tok"])
# self.gguf_writer.add_expert_count(self.hparams["num_local_experts"])

def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
# process the experts separately
if name.find("block_sparse_moe.experts") != -1:
n_experts = self.hparams["num_local_experts"]
assert bid is not None
# def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
# # process the experts separately
# if name.find("block_sparse_moe.experts") != -1:
# n_experts = self.hparams["num_local_experts"]
# assert bid is not None

if self._experts is None:
self._experts = [{} for _ in range(self.block_count)]
# if self._experts is None:
# self._experts = [{} for _ in range(self.block_count)]

self._experts[bid][name] = data_torch
# self._experts[bid][name] = data_torch

if len(self._experts[bid]) >= n_experts * 3:
tensors: list[tuple[str, Tensor]] = []
# if len(self._experts[bid]) >= n_experts * 3:
# tensors: list[tuple[str, Tensor]] = []

# merge the experts into a single 3d tensor
for w_name in ["w1", "w2", "w3"]:
data: list[Tensor] = []
# # merge the experts into a single 3d tensor
# for w_name in ["w1", "w2", "w3"]:
# data: list[Tensor] = []

for xid in range(n_experts):
ename = f"model.layers.{bid}.block_sparse_moe.experts.{xid}.{w_name}.weight"
data.append(self._experts[bid][ename])
del self._experts[bid][ename]
# for xid in range(n_experts):
# ename = f"model.layers.{bid}.block_sparse_moe.experts.{xid}.{w_name}.weight"
# data.append(self._experts[bid][ename])
# del self._experts[bid][ename]

data_torch = torch.stack(data, dim=0)
# data_torch = torch.stack(data, dim=0)

merged_name = f"model.layers.{bid}.block_sparse_moe.experts.{w_name}.weight"
# merged_name = f"model.layers.{bid}.block_sparse_moe.experts.{w_name}.weight"

new_name = self.map_tensor_name(merged_name)
# new_name = self.map_tensor_name(merged_name)

tensors.append((new_name, data_torch))
return tensors
else:
return []
# tensors.append((new_name, data_torch))
# return tensors
# else:
# return []

return [(self.map_tensor_name(name), data_torch)]
# return [(self.map_tensor_name(name), data_torch)]

def prepare_tensors(self):
super().prepare_tensors()
# def prepare_tensors(self):
# super().prepare_tensors()

if self._experts is not None:
# flatten `list[dict[str, Tensor]]` into `list[str]`
experts = [k for d in self._experts for k in d.keys()]
if len(experts) > 0:
raise ValueError(f"Unprocessed experts: {experts}")
# if self._experts is not None:
# # flatten `list[dict[str, Tensor]]` into `list[str]`
# experts = [k for d in self._experts for k in d.keys()]
# if len(experts) > 0:
# raise ValueError(f"Unprocessed experts: {experts}")


@Model.register("PlamoForCausalLM")
Expand Down Expand Up @@ -3667,7 +3711,7 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
if name.startswith("language_model."):
name = name.replace("language_model.", "")
elif name.startswith("multi_modal_projector.") or name.startswith("vision_tower.") \
or name.startswith("multimodal_projector.") or name.startswith("vision_model."):
or name.startswith("multimodal_projector.") or name.startswith("vision_model."):
# ignore vision tensors
return []

Expand Down
1 change: 1 addition & 0 deletions auto_round/export/export_to_gguf/quant.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -303,6 +303,7 @@ def q2_k_quant_block(blocks: np.array, scale=None, zp=None, wmin_m=None, d_scale

@register_block("q4_k")
def q4_k_quant_block(blocks: np.array, scale=None, zp=None, wmin_m=None, d_scale=None, d_wmin_m=None):
# print(f"q4_k_quant_block: {blocks.shape=}, {blocks.dtype=}, {scale.shape=}, {scale.dtype=}, {wmin_m.shape=}, {wmin_m.dtype=}, {d_scale.shape=}, {d_scale.dtype=}, {d_wmin_m.shape=} {d_wmin_m.dtype=}")
nb = blocks.shape[0]
output_scale = np.empty((nb, K_SCALE_SIZE), dtype=np.uint8)
output_d = np.empty(nb, dtype=np.float32)
Expand Down
Loading