Skip to content

Commit

Permalink
convert : refactor rope_freqs generation (ggerganov#9396)
Browse files Browse the repository at this point in the history
* convert : refactor rope_freqs generation

This should also fix vocab-only conversion for Phi-3.

* convert : adapt MiniCPM3 to separate rope_freqs insertion

MiniCPM3's tokenizer is treated as a SentencePiece tokenizer to avoid
having to run its custom Python code which mixes tokenization
in the same file as tool calls.

gguf-py : add long and short RoPE factors to tensor mappings

Empty, but the key names are used to populate the mappings.
  • Loading branch information
compilade authored and arthw committed Nov 15, 2024
1 parent bea0155 commit 8fe8208
Show file tree
Hide file tree
Showing 4 changed files with 44 additions and 29 deletions.
61 changes: 33 additions & 28 deletions convert_hf_to_gguf.py
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@
from pathlib import Path
from hashlib import sha256
from typing import TYPE_CHECKING, Any, Callable, ContextManager, Iterable, Iterator, Literal, Sequence, TypeVar, cast
from itertools import chain

import math
import numpy as np
Expand Down Expand Up @@ -64,15 +65,14 @@ class Model:
model_name: str | None
metadata_override: Path | None
dir_model_card: Path
is_lora: bool

# subclasses should define this!
model_arch: gguf.MODEL_ARCH

def __init__(self, dir_model: Path, ftype: gguf.LlamaFileType, fname_out: Path, is_big_endian: bool = False,
use_temp_file: bool = False, eager: bool = False,
metadata_override: Path | None = None, model_name: str | None = None,
split_max_tensors: int = 0, split_max_size: int = 0, dry_run: bool = False, small_first_shard: bool = False, is_lora: bool = False):
split_max_tensors: int = 0, split_max_size: int = 0, dry_run: bool = False, small_first_shard: bool = False):
if type(self) is Model:
raise TypeError(f"{type(self).__name__!r} should not be directly instantiated")

Expand All @@ -94,7 +94,6 @@ def __init__(self, dir_model: Path, ftype: gguf.LlamaFileType, fname_out: Path,
self.metadata_override = metadata_override
self.model_name = model_name
self.dir_model_card = dir_model # overridden in convert_lora_to_gguf.py
self.is_lora = is_lora # true if model is used inside convert_lora_to_gguf.py

# Apply heuristics to figure out typical tensor encoding based on first layer tensor encoding type
if self.ftype == gguf.LlamaFileType.GUESSED:
Expand Down Expand Up @@ -270,10 +269,14 @@ def tensor_force_quant(self, name: str, new_name: str, bid: int | None, n_dims:

return False

# some models need extra generated tensors (like rope_freqs)
def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
return ()

def prepare_tensors(self):
max_name_len = max(len(s) for _, s in self.tensor_map.mapping.values()) + len(".weight,")

for name, data_torch in self.get_tensors():
for name, data_torch in chain(self.generate_extra_tensors(), self.get_tensors()):
# we don't need these
if name.endswith((".attention.masked_bias", ".attention.bias", ".rotary_emb.inv_freq")):
continue
Expand Down Expand Up @@ -1617,7 +1620,7 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter

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

def prepare_tensors(self):
def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
if rope_scaling := self.find_hparam(["rope_scaling"], optional=True):
if rope_scaling.get("rope_type", '').lower() == "llama3":
base = self.hparams.get("rope_theta", 10000.0)
Expand All @@ -1644,9 +1647,9 @@ def prepare_tensors(self):
smooth = (old_context_len / wavelen - low_freq_factor) / (high_freq_factor - low_freq_factor)
rope_factors.append(1 / ((1 - smooth) / factor + smooth))

if not self.is_lora:
self.gguf_writer.add_tensor(self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FREQS), np.array(rope_factors, dtype=np.float32))
yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FREQS), torch.tensor(rope_factors, dtype=torch.float32))

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

if self._experts is not None:
Expand Down Expand Up @@ -1870,8 +1873,6 @@ class MiniCPM3Model(Model):
def set_gguf_parameters(self):
hparams = self.hparams

rope_dims = hparams["qk_rope_head_dim"]

self.gguf_writer.add_file_type(self.ftype)
self.gguf_writer.add_context_length(hparams["max_position_embeddings"])
self.gguf_writer.add_embedding_length(hparams["hidden_size"])
Expand All @@ -1887,24 +1888,25 @@ def set_gguf_parameters(self):
self.gguf_writer.add_key_length(hparams["qk_nope_head_dim"] + hparams["qk_rope_head_dim"])
self.gguf_writer.add_rope_dimension_count(hparams["qk_rope_head_dim"])

def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
rope_scaling = self.find_hparam(['rope_scaling'], True)
if rope_scaling is None:
return
if rope_scaling is not None:
rope_dims = self.hparams["qk_rope_head_dim"]

long_factors = rope_scaling.get('long_factor', None)
short_factors = rope_scaling.get('short_factor', None)
long_factors = rope_scaling.get('long_factor', None)
short_factors = rope_scaling.get('short_factor', None)

if long_factors is None or short_factors is None:
raise KeyError('Missing the required key rope_scaling.long_factor or rope_scaling_short_factor')
if long_factors is None or short_factors is None:
raise KeyError('Missing the required key rope_scaling.long_factor or rope_scaling_short_factor')

if len(long_factors) != len(short_factors) or len(long_factors) != rope_dims / 2:
raise ValueError(f'The length of rope long and short factors must be {rope_dims / 2}')
if len(long_factors) != len(short_factors) or len(long_factors) != rope_dims / 2:
raise ValueError(f'The length of rope long and short factors must be {rope_dims / 2}')

self.gguf_writer.add_tensor(gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.ROPE_FACTORS_LONG] + ".weight", np.array(long_factors, dtype=np.float32))
self.gguf_writer.add_tensor(gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.ROPE_FACTORS_SHORT] + ".weight", np.array(short_factors, dtype=np.float32))
yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FACTORS_LONG), torch.tensor(long_factors, dtype=torch.float32))
yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FACTORS_SHORT), torch.tensor(short_factors, dtype=torch.float32))

def set_vocab(self):
self._set_vocab_llama_hf()
self._set_vocab_sentencepiece()

def _reverse_hf_permute(self, weights: Tensor, n_head: int, n_kv_head: int | None = None) -> Tensor:
if n_kv_head is not None and n_head != n_kv_head:
Expand Down Expand Up @@ -2216,6 +2218,13 @@ def set_gguf_parameters(self):
self.gguf_writer.add_file_type(self.ftype)
self.gguf_writer.add_sliding_window(self.find_hparam(["sliding_window"]))

def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
n_embd = self.find_hparam(["hidden_size", "n_embd"])
n_head = self.find_hparam(["num_attention_heads", "n_head"])
max_pos_embds = self.find_hparam(["n_positions", "max_position_embeddings"])
orig_max_pos_embds = self.find_hparam(["original_max_position_embeddings"])
rope_dims = n_embd // n_head

# write rope scaling for long context (128k) model
rope_scaling = self.find_hparam(['rope_scaling'], True)
if rope_scaling is None:
Expand Down Expand Up @@ -2245,9 +2254,8 @@ def set_gguf_parameters(self):
if len(long_factors) != len(short_factors) or len(long_factors) != rope_dims / 2:
raise ValueError(f'The length of rope long and short factors must be {rope_dims / 2}')

if not self.is_lora:
self.gguf_writer.add_tensor(gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.ROPE_FACTORS_LONG] + ".weight", np.array(long_factors, dtype=np.float32))
self.gguf_writer.add_tensor(gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.ROPE_FACTORS_SHORT] + ".weight", np.array(short_factors, dtype=np.float32))
yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FACTORS_LONG), torch.tensor(long_factors, dtype=torch.float32))
yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FACTORS_SHORT), torch.tensor(short_factors, dtype=torch.float32))


@Model.register("PlamoForCausalLM")
Expand Down Expand Up @@ -4071,7 +4079,7 @@ def set_gguf_parameters(self):
self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.LINEAR)
self.gguf_writer.add_rope_scaling_factor(hparams["rope_scaling"]["factor"])

def prepare_tensors(self):
def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
if rope_scaling := self.find_hparam(["rope_scaling"], optional=True):
if rope_scaling.get("rope_type", '').lower() == "llama3":
base = self.hparams.get("rope_theta", 10000.0)
Expand All @@ -4098,10 +4106,7 @@ def prepare_tensors(self):
smooth = (old_context_len / wavelen - low_freq_factor) / (high_freq_factor - low_freq_factor)
rope_factors.append(1 / ((1 - smooth) / factor + smooth))

if not self.is_lora:
self.gguf_writer.add_tensor(self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FREQS), np.array(rope_factors, dtype=np.float32))

super().prepare_tensors()
yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FREQS), torch.tensor(rope_factors, dtype=torch.float32))


@Model.register("GraniteForCausalLM")
Expand Down
5 changes: 4 additions & 1 deletion convert_lora_to_gguf.py
Original file line number Diff line number Diff line change
Expand Up @@ -331,6 +331,10 @@ def set_gguf_parameters(self):
self.gguf_writer.add_float32(gguf.Keys.Adapter.LORA_ALPHA, self.lora_alpha)
super().set_gguf_parameters()

def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
# Never add extra tensors (e.g. rope_freqs) for LoRA adapters
return ()

def get_tensors(self) -> Iterator[tuple[str, Tensor]]:
tensor_map: dict[str, PartialLoraTensor] = {}

Expand Down Expand Up @@ -392,7 +396,6 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
dry_run=args.dry_run,
dir_lora_model=dir_lora,
lora_alpha=alpha,
is_lora=True,
)

logger.info("Exporting model...")
Expand Down
4 changes: 4 additions & 0 deletions gguf-py/gguf/constants.py
Original file line number Diff line number Diff line change
Expand Up @@ -814,6 +814,8 @@ class MODEL_TENSOR(IntEnum):
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ROPE_FACTORS_LONG,
MODEL_TENSOR.ROPE_FACTORS_SHORT,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_QKV,
MODEL_TENSOR.ATTN_Q,
Expand Down Expand Up @@ -892,6 +894,8 @@ class MODEL_TENSOR(IntEnum):
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ROPE_FACTORS_LONG,
MODEL_TENSOR.ROPE_FACTORS_SHORT,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q_A,
MODEL_TENSOR.ATTN_Q_B,
Expand Down
3 changes: 3 additions & 0 deletions gguf-py/gguf/tensor_mapping.py
Original file line number Diff line number Diff line change
Expand Up @@ -87,6 +87,9 @@ class TensorNameMap:
"rope.freqs", # llama-pth
"rotary_pos_emb.inv_freq", # chatglm
),

MODEL_TENSOR.ROPE_FACTORS_LONG: (),
MODEL_TENSOR.ROPE_FACTORS_SHORT: (),
}

block_mappings_cfg: dict[MODEL_TENSOR, tuple[str, ...]] = {
Expand Down

0 comments on commit 8fe8208

Please sign in to comment.