Files
Xiaoze Fan 477c8601b2 refactor(quant): config, scheme and method layers for quantization (#418)
* refactor(quant): config, scheme and method layers for checkpoint quantization

QuantConfig parses the checkpoint's quantization_config into a QuantScheme per
projection; every Linear*, ParallelLMHead and MoE layer takes quant_config and a
prefix, declares its tensors through a QuantMethod and selects its kernel at
construction. MoE kernels own the expert bank layout, pack and apply, so the
offload cache, the FTW converter and the CPU executor follow one layout.

- --quant-backend layer[.kind]=name selects kernels (--nvfp4-backend forwards to it)
- --moe-strategy replaces --moe-backend; the old spelling stays as an alias
- --moe-cpu-layers auto for Windows/WSL pin budgets; --moe-cache-auto respects the kernel's slot limit
- removes the per-family quantized layer classes, bank providers, format env switches and GLM runtime fp8
2026-09-08 23:37:23 -07:00

92 lines
3.1 KiB
Python

from __future__ import annotations
from typing import TYPE_CHECKING, Tuple
import torch
from freetoken.core import get_global_ctx
from freetoken.layers import BaseOP, OPList, ParallelLMHead, RMSNormFused, VocabParallelEmbedding
from freetoken.utils import nvtx_annotate
from freetoken.models.blocks import BaseLLMModel, GatedMLP as MistralMLP
from .attention import MistralAttention as MistralAttn
if TYPE_CHECKING:
from freetoken.models.config import ModelConfig
class MistralDecoderLayer(BaseOP):
def __init__(self, config: ModelConfig, layer_id: int, *, prefix: str = ""):
self.self_attn = MistralAttn(config, layer_id, prefix=f"{prefix}.self_attn")
self.mlp = MistralMLP(config, quant_config=config.quant, prefix=f"{prefix}.mlp")
self.input_layernorm = RMSNormFused(
size=config.hidden_size,
eps=config.rms_norm_eps,
)
self.post_attention_layernorm = RMSNormFused(
size=config.hidden_size,
eps=config.rms_norm_eps,
)
self._layer_id = layer_id
@nvtx_annotate("Layer_{}", layer_id_field="_layer_id")
def forward(
self,
x: torch.Tensor,
residual: torch.Tensor | None = None,
) -> Tuple[torch.Tensor, torch.Tensor]:
x, residual = self.input_layernorm.forward(x, residual)
x = self.self_attn.forward(x)
x, residual = self.post_attention_layernorm.forward(x, residual)
x = self.mlp.forward(x)
return x, residual
class MistralModel(BaseOP):
def __init__(self, config: ModelConfig, *, prefix: str = "model"):
self.embed_tokens = VocabParallelEmbedding(
num_embeddings=config.vocab_size,
embedding_dim=config.hidden_size,
)
self.layers = OPList(
[
MistralDecoderLayer(config, layer_id, prefix=f"{prefix}.layers.{layer_id}")
for layer_id in range(config.num_layers)
]
)
self.norm = RMSNormFused(
size=config.hidden_size,
eps=config.rms_norm_eps,
)
def forward(self, input_ids: torch.Tensor) -> torch.Tensor:
x = self.embed_tokens.forward(input_ids)
residual: torch.Tensor | None = None
for layer in self.layers.op_list:
x, residual = layer.forward(x, residual)
return self.norm.forward(x, residual)[0]
class MistralForCausalLM(BaseLLMModel):
def __init__(self, config: ModelConfig):
self.model = MistralModel(config)
self.lm_head = ParallelLMHead(
num_embeddings=config.vocab_size,
embedding_dim=config.hidden_size,
tie_word_embeddings=config.tie_word_embeddings,
tied_embedding=self.model.embed_tokens if config.tie_word_embeddings else None,
quant_config=config.quant,
prefix="lm_head",
)
super().__init__()
def forward(self) -> torch.Tensor:
ids = get_global_ctx().batch.input_ids
output = self.model.forward(ids)
logits = self.lm_head.forward(output)
return logits
__all__ = ["MistralForCausalLM"]