CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama : add StableLM2 12B (#6635)

Browse Source 
* StableLM2 12B support for huggingface -> GGUF

* StableLM12 tensormapping and constants

* StableLM-2-12b model support

* fix

* Added 12B support

* Removed autoformatting; resolved bug where model_arch was not selecting StableLM2

* Formatting

* Do QK norm stacking in model conversion step

* Converge StableLM and StableLM2 code to simplify graph construction

* Fix accidental removal

* Removed warnings

* Revert formatter

* Move QK norm stack to private function so it's easier to read

* refactor stablelm graph builder to support 1.6, 3b and 12b more efficiently

* Proper check for None type for new_name to avoid crash; formatting; revert change to base class `write_tensors()`

* Format

* Formatting

* format

Co-authored-by: compilade <git@compilade.net>

* Fix incorrect check for K norm

* space after commas; Keep indentation multiple of 4 spaces

* Flake8 format

* Removed unnecessary conditional branches

* Removed unused comment

* Fixed incorrect tensor passing

* Format

---------

Co-authored-by: compilade <git@compilade.net>
This commit is contained in:
Ashish
2024-04-16 08:48:35 -07:00
committed by GitHub
parent f4dea7da18
commit dbceec87c0
3 changed files with 134 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

82
convert-hf-to-gguf.py
View File

@@ -1207,9 +1207,91 @@ class StableLMModel(Model):
rotary_factor = self.find_hparam(["partial_rotary_factor", "rope_pct"])
self.gguf_writer.add_rope_dimension_count(int(rotary_factor * (hparams["hidden_size"] // hparams["num_attention_heads"])))
self.gguf_writer.add_head_count(hparams["num_attention_heads"])
self.gguf_writer.add_head_count_kv(hparams["num_key_value_heads"])
self.gguf_writer.add_parallel_residual(hparams["use_parallel_residual"] if "use_parallel_residual" in hparams else True)
self.gguf_writer.add_layer_norm_eps(self.find_hparam(["layer_norm_eps", "norm_eps"]))
def write_tensors(self):
block_count = self.hparams.get("n_layers", self.hparams.get("num_hidden_layers", self.hparams.get("n_layer")))
tensor_map = gguf.get_tensor_name_map(self.model_arch, block_count)
n_head = self.hparams.get("num_attention_heads")
n_kv_head = self.hparams.get("num_key_value_heads")
q_norms = dict()
k_norms = dict()
for name, data_torch in self.get_tensors():
# we don't need these
if name.endswith((".attention.masked_bias", ".attention.bias", ".attention.rotary_emb.inv_freq")):
continue
old_dtype = data_torch.dtype
# convert any unsupported data types to float32
if data_torch.dtype not in (torch.float16, torch.float32):
data_torch = data_torch.to(torch.float32)
data = data_torch.squeeze().numpy()
n_dims = len(data.shape)
if name.find("q_layernorm.norms") != -1:
q_norms[name] = data
if len(q_norms) >= (block_count * n_head):
self._stack_qk_norm(block_count, name, tensor_map, n_head, q_norms, n_dims, layer_name="q_layernorm")
continue
if name.find("k_layernorm.norms") != -1:
k_norms[name] = data
if len(k_norms) >= (block_count * n_kv_head):
self._stack_qk_norm(block_count, name, tensor_map, n_kv_head, k_norms, n_dims, layer_name="k_layernorm")
continue
# map tensor names
new_name = tensor_map.get_name(name, try_suffixes=(".weight", ".bias"))
if new_name is None:
print(f"Can not map tensor {name!r}")
sys.exit()
n_dims = len(data.shape)
data_dtype = data.dtype
# if f32 desired, convert any float16 to float32
if self.ftype == 0 and data_dtype == np.float16:
data = data.astype(np.float32)
# TODO: Why cant we use these float16 as-is? There should be not reason to store float16 as float32
if self.ftype == 1 and data_dtype == np.float16 and (n_dims == 1 or new_name.endswith("_norm.weight")):
data = data.astype(np.float32)
# if f16 desired, convert any float32 2-dim weight tensors to float16
if self.ftype == 1 and data_dtype == np.float32 and name.endswith(".weight") and not new_name.endswith("_norm.weight") and n_dims == 2:
data = data.astype(np.float16)
print(f"{new_name}, n_dims = {n_dims}, {old_dtype} --> {data.dtype}")
self.gguf_writer.add_tensor(new_name, data)
def _stack_qk_norm(self, block_count, name, tensor_map, n_head, norms, n_dims, layer_name="q_layernorm"):
for bid in range(block_count):
datas = []
for xid in range(n_head):
ename = f"model.layers.{bid}.self_attn.{layer_name}.norms.{xid}.weight"
datas.append(norms[ename])
del norms[ename]
data = np.stack(datas, axis=0)
data_dtype = data.dtype
merged_name = f"model.layers.{bid}.self_attn.{layer_name}.weight"
new_name = tensor_map.get_name(merged_name, try_suffixes=(".weight", ".bias"))
if new_name is None:
print(f"Can not map tensor {name!r}")
sys.exit()
if self.ftype == 1 and data_dtype == np.float16 and (n_dims == 1 or new_name.endswith("_norm.weight")):
data = data.astype(np.float32)
# if f16 desired, convert any float32 2-dim weight tensors to float16
if self.ftype == 1 and data_dtype == np.float32 and name.endswith(".weight") and not new_name.endswith("_norm.weight") and n_dims == 2:
data = data.astype(np.float16)
print(f"{new_name}, n_dims = {len(data.shape)}, shape = {data.shape} --> {data.dtype}")
self.gguf_writer.add_tensor(new_name, data)
@Model.register("LlamaForCausalLM", "MistralForCausalLM", "MixtralForCausalLM")
class LlamaModel(Model):