CodeCow/stable-diffusion-webui-forge
1
0
Fork 0
mirror of https://github.com/lllyasviel/stable-diffusion-webui-forge.git synced 2026-07-21 21:01:24 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

emphasis for LLM

Browse Source
This commit is contained in:
Haoming 2026-02-01 00:26:22 +08:00
parent 78af5c4d44
commit 527db2e0fd
4 changed files with 30 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
backend/text_processing/gemma_engine.py
View File

@ -43,10 +43,8 @@ class GemmaTextProcessingEngine:
chunks = []
chunk = PromptChunk()
token_count = 0
def next_chunk():
nonlocal token_count
nonlocal chunk
chunk.tokens = [self.id_start] + chunk.tokens
@ -70,7 +68,7 @@ class GemmaTextProcessingEngine:
if chunk.tokens or not chunks:
next_chunk()
return chunks, token_count
return chunks
@staticmethod
def process_template(text: str, negative: bool) -> str:
@ -96,23 +94,13 @@ class GemmaTextProcessingEngine:
if line in cache:
line_z_values = cache[line]
else:
chunks, token_count = self.tokenize_line(line)
chunks = self.tokenize_line(line)
line_z_values = []
# pad all chunks to length of longest chunk
# max_tokens = 0
# for chunk in chunks:
# max_tokens = max(len(chunk.tokens), max_tokens)
for chunk in chunks:
tokens = chunk.tokens
multipliers = chunk.multipliers
# remaining_count = max_tokens - len(tokens)
# if remaining_count > 0:
# tokens += [self.id_pad] * remaining_count
# multipliers += [1.0] * remaining_count
z = self.process_tokens([tokens], [multipliers])[0]
line_z_values.append(z)
cache[line] = line_z_values
@ -155,6 +143,13 @@ class GemmaTextProcessingEngine:
def process_tokens(self, batch_tokens, batch_multipliers):
embeds, mask, count = self.process_embeds(batch_tokens)
self.emphasis.tokens = batch_tokens
self.emphasis.multipliers = torch.asarray(batch_multipliers).to(embeds)
self.emphasis.z = embeds
self.emphasis.after_transformers()
embeds = self.emphasis.z
_, z = self.text_encoder(
None,
attention_mask=mask,

7
backend/text_processing/klein_engine.py
View File

@ -128,6 +128,13 @@ class KleinTextProcessingEngine:
def process_tokens(self, batch_tokens, batch_multipliers):
embeds, mask, count = self.process_embeds(batch_tokens)
self.emphasis.tokens = batch_tokens
self.emphasis.multipliers = torch.asarray(batch_multipliers).to(embeds)
self.emphasis.z = embeds
self.emphasis.after_transformers()
embeds = self.emphasis.z
_, out = self.text_encoder(
None,
attention_mask=mask,

7
backend/text_processing/qwen3_engine.py
View File

@ -120,6 +120,13 @@ class Qwen3TextProcessingEngine:
def process_tokens(self, batch_tokens, batch_multipliers):
embeds, mask, count = self.process_embeds(batch_tokens)
self.emphasis.tokens = batch_tokens
self.emphasis.multipliers = torch.asarray(batch_multipliers).to(embeds)
self.emphasis.z = embeds
self.emphasis.after_transformers()
embeds = self.emphasis.z
_, z = self.text_encoder(
None,
attention_mask=mask,

7
backend/text_processing/qwen_engine.py
View File

@ -175,5 +175,12 @@ class QwenTextProcessingEngine:
def process_tokens(self, batch_tokens, batch_multipliers):
embeds, mask, count, info = self.process_embeds(batch_tokens)
self.emphasis.tokens = batch_tokens
self.emphasis.multipliers = torch.asarray(batch_multipliers).to(embeds)
self.emphasis.z = embeds
self.emphasis.after_transformers()
embeds = self.emphasis.z
z, _ = self.text_encoder(x=None, embeds=embeds, attention_mask=mask, num_tokens=count, embeds_info=info)
return z