vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

convert.py: named instance logging

Browse source
This commit is contained in:
brian khuu 2024-04-08 10:19:46 +10:00
parent 88c1e2ff10
commit e8be0c8f73
1 changed files with 13 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

24
convert.py
View file

@ -36,6 +36,8 @@ import gguf
if TYPE_CHECKING: if TYPE_CHECKING:
from typing_extensions import Self, TypeAlias from typing_extensions import Self, TypeAlias
logger = logging.getLogger(__name__)
if hasattr(faulthandler, 'register') and hasattr(signal, 'SIGUSR1'): if hasattr(faulthandler, 'register') and hasattr(signal, 'SIGUSR1'):
faulthandler.register(signal.SIGUSR1) faulthandler.register(signal.SIGUSR1)
@ -644,7 +646,7 @@ class LlamaHfVocab(Vocab):
def permute(weights: NDArray, n_head: int, n_head_kv: int) -> NDArray: def permute(weights: NDArray, n_head: int, n_head_kv: int) -> NDArray:
# logging.info( "permute debug " + str(weights.shape[0]) + " x " + str(weights.shape[1]) + " nhead " + str(n_head) + " nheadkv " + str(n_kv_head) ) # logger.info( "permute debug " + str(weights.shape[0]) + " x " + str(weights.shape[1]) + " nhead " + str(n_head) + " nheadkv " + str(n_kv_head) )
if n_head_kv is not None and n_head != n_head_kv: if n_head_kv is not None and n_head != n_head_kv:
n_head = n_head_kv n_head = n_head_kv
return (weights.reshape(n_head, 2, weights.shape[0] // n_head // 2, *weights.shape[1:]) return (weights.reshape(n_head, 2, weights.shape[0] // n_head // 2, *weights.shape[1:])
@ -1034,12 +1036,12 @@ def check_vocab_size(params: Params, vocab: BaseVocab, pad_vocab: bool = False)
# Check for a vocab size mismatch # Check for a vocab size mismatch
if params.n_vocab == vocab.vocab_size: if params.n_vocab == vocab.vocab_size:
logging.warning("Ignoring added_tokens.json since model matches vocab size without it.") logger.warning("Ignoring added_tokens.json since model matches vocab size without it.")
return return
if pad_vocab and params.n_vocab > vocab.vocab_size: if pad_vocab and params.n_vocab > vocab.vocab_size:
pad_count = params.n_vocab - vocab.vocab_size pad_count = params.n_vocab - vocab.vocab_size
logging.debug( logger.debug(
f"Padding vocab with {pad_count} token(s) - <dummy00001> through <dummy{pad_count:05}>" f"Padding vocab with {pad_count} token(s) - <dummy00001> through <dummy{pad_count:05}>"
) )
for i in range(1, pad_count + 1): for i in range(1, pad_count + 1):
@ -1167,7 +1169,7 @@ class OutputFile:
elapsed = time.time() - start elapsed = time.time() - start
size = ' x '.join(f"{dim:6d}" for dim in lazy_tensor.shape) size = ' x '.join(f"{dim:6d}" for dim in lazy_tensor.shape)
padi = len(str(len(model))) padi = len(str(len(model)))
logging.info( logger.info(
f"[{i + 1:{padi}d}/{len(model)}] Writing tensor {name:38s} | size {size:16} | type {lazy_tensor.data_type.name:4} | T+{int(elapsed):4}" f"[{i + 1:{padi}d}/{len(model)}] Writing tensor {name:38s} | size {size:16} | type {lazy_tensor.data_type.name:4} | T+{int(elapsed):4}"
) )
self.gguf.write_tensor_data(ndarray) self.gguf.write_tensor_data(ndarray)
@ -1282,12 +1284,12 @@ def convert_model_names(model: LazyModel, params: Params, skip_unknown: bool) ->
# HF models permut or pack some of the tensors, so we need to undo that # HF models permut or pack some of the tensors, so we need to undo that
for i in itertools.count(): for i in itertools.count():
if f"model.layers.{i}.self_attn.q_proj.weight" in model: if f"model.layers.{i}.self_attn.q_proj.weight" in model:
logging.debug(f"Permuting layer {i}") logger.debug(f"Permuting layer {i}")
tmp[f"model.layers.{i}.self_attn.q_proj.weight"] = permute_lazy(model[f"model.layers.{i}.self_attn.q_proj.weight"], params.n_head, params.n_head) tmp[f"model.layers.{i}.self_attn.q_proj.weight"] = permute_lazy(model[f"model.layers.{i}.self_attn.q_proj.weight"], params.n_head, params.n_head)
tmp[f"model.layers.{i}.self_attn.k_proj.weight"] = permute_lazy(model[f"model.layers.{i}.self_attn.k_proj.weight"], params.n_head, params.n_head_kv) tmp[f"model.layers.{i}.self_attn.k_proj.weight"] = permute_lazy(model[f"model.layers.{i}.self_attn.k_proj.weight"], params.n_head, params.n_head_kv)
# tmp[f"model.layers.{i}.self_attn.v_proj.weight"] = model[f"model.layers.{i}.self_attn.v_proj.weight"] # tmp[f"model.layers.{i}.self_attn.v_proj.weight"] = model[f"model.layers.{i}.self_attn.v_proj.weight"]
elif f"model.layers.{i}.self_attn.W_pack.weight" in model: elif f"model.layers.{i}.self_attn.W_pack.weight" in model:
logging.debug(f"Unpacking and permuting layer {i}") logger.debug(f"Unpacking and permuting layer {i}")
tmp[f"model.layers.{i}.self_attn.q_proj.weight"] = permute_part_lazy(model[f"model.layers.{i}.self_attn.W_pack.weight"], 0, params.n_head, params.n_head) tmp[f"model.layers.{i}.self_attn.q_proj.weight"] = permute_part_lazy(model[f"model.layers.{i}.self_attn.W_pack.weight"], 0, params.n_head, params.n_head)
tmp[f"model.layers.{i}.self_attn.k_proj.weight"] = permute_part_lazy(model[f"model.layers.{i}.self_attn.W_pack.weight"], 1, params.n_head, params.n_head_kv) tmp[f"model.layers.{i}.self_attn.k_proj.weight"] = permute_part_lazy(model[f"model.layers.{i}.self_attn.W_pack.weight"], 1, params.n_head, params.n_head_kv)
tmp[f"model.layers.{i}.self_attn.v_proj.weight"] = part_lazy (model[f"model.layers.{i}.self_attn.W_pack.weight"], 2) tmp[f"model.layers.{i}.self_attn.v_proj.weight"] = part_lazy (model[f"model.layers.{i}.self_attn.W_pack.weight"], 2)
@ -1300,15 +1302,15 @@ def convert_model_names(model: LazyModel, params: Params, skip_unknown: bool) ->
tensor_type, name_new = tmap.get_type_and_name(name, try_suffixes = (".weight", ".bias")) or (None, None) tensor_type, name_new = tmap.get_type_and_name(name, try_suffixes = (".weight", ".bias")) or (None, None)
if name_new is None: if name_new is None:
if skip_unknown: if skip_unknown:
logging.warning(f"Unexpected tensor name: {name} - skipping") logger.warning(f"Unexpected tensor name: {name} - skipping")
continue continue
raise ValueError(f"Unexpected tensor name: {name}. Use --skip-unknown to ignore it (e.g. LLaVA)") raise ValueError(f"Unexpected tensor name: {name}. Use --skip-unknown to ignore it (e.g. LLaVA)")
if tensor_type in should_skip: if tensor_type in should_skip:
logging.debug(f"skipping tensor {name_new}") logger.debug(f"skipping tensor {name_new}")
continue continue
logging.debug(f"{name:48s} -> {name_new:40s} | {lazy_tensor.data_type.name:6s} | {lazy_tensor.shape}") logger.debug(f"{name:48s} -> {name_new:40s} | {lazy_tensor.data_type.name:6s} | {lazy_tensor.shape}")
out[name_new] = lazy_tensor out[name_new] = lazy_tensor
return out return out
@ -1373,7 +1375,7 @@ def load_some_model(path: Path) -> ModelPlus:
paths = find_multifile_paths(path) paths = find_multifile_paths(path)
models_plus: list[ModelPlus] = [] models_plus: list[ModelPlus] = []
for path in paths: for path in paths:
logging.info(f"Loading model file {path}") logger.info(f"Loading model file {path}")
models_plus.append(lazy_load_file(path)) models_plus.append(lazy_load_file(path))
model_plus = merge_multifile_models(models_plus) model_plus = merge_multifile_models(models_plus)
@ -1414,7 +1416,7 @@ class VocabFactory:
else: else:
raise FileNotFoundError(f"Could not find a tokenizer matching any of {vocab_types}") raise FileNotFoundError(f"Could not find a tokenizer matching any of {vocab_types}")
logging.info(f"Loaded vocab file {vocab.fname_tokenizer!r}, type {vocab.name!r}") logger.info(f"Loaded vocab file {vocab.fname_tokenizer!r}, type {vocab.name!r}")
return vocab return vocab
def load_vocab(self, vocab_types: list[str] | None, model_parent_path: Path) -> tuple[BaseVocab, gguf.SpecialVocab]: def load_vocab(self, vocab_types: list[str] | None, model_parent_path: Path) -> tuple[BaseVocab, gguf.SpecialVocab]: