rixty/KoboldAI-Client
1
1
Fork 0
mirror of https://github.com/KoboldAI/KoboldAI-Client.git synced 2025-06-05 21:59:24 +02:00
Code Issues Projects Releases Wiki Activity

Finally some debugging

Browse Source
This commit is contained in:
somebody
2022-09-22 19:28:00 -05:00
parent de6ea04e12
commit 3ddf6170f6
1 changed files with 10 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
aiserver.py
View File

@@ -4839,6 +4839,9 @@ def calcsubmit(txt):
# Send it! # Send it!
ikrequest(subtxt) ikrequest(subtxt)
def __debug(text):
print(f"[DBG] {text}")
def core_generate(text: list, min: int, max: int, found_entries: set): def core_generate(text: list, min: int, max: int, found_entries: set):
# This generation function is tangled with koboldai_vars intentionally. It # This generation function is tangled with koboldai_vars intentionally. It
# is meant for the story and nothing else. # is meant for the story and nothing else.
@@ -4868,6 +4871,7 @@ def core_generate(text: list, min: int, max: int, found_entries: set):
koboldai_vars._prompt = koboldai_vars.prompt koboldai_vars._prompt = koboldai_vars.prompt
__debug("generate core", text)
with torch.no_grad(): with torch.no_grad():
already_generated = 0 already_generated = 0
numseqs = koboldai_vars.numseqs numseqs = koboldai_vars.numseqs
@@ -4875,6 +4879,7 @@ def core_generate(text: list, min: int, max: int, found_entries: set):
do_loop = True do_loop = True
while do_loop: while do_loop:
__debug("generate loop start", text)
# The reason this is a loop is due to how Dynamic WI works. We # The reason this is a loop is due to how Dynamic WI works. We
# cannot simply add the WI to the context mid-generation, so we # cannot simply add the WI to the context mid-generation, so we
# stop early, and then insert WI, then continue generating. That # stop early, and then insert WI, then continue generating. That
@@ -4890,7 +4895,9 @@ def core_generate(text: list, min: int, max: int, found_entries: set):
bypass_hf_maxlength=True, bypass_hf_maxlength=True,
) )
__debug("generate result", result.__dict__)
do_loop = not result.is_whole_generation do_loop = not result.is_whole_generation
__debug("loop is", do_loop)
genout = result.encoded genout = result.encoded
already_generated += len(genout[0]) - len(gen_in[0]) already_generated += len(genout[0]) - len(gen_in[0])
@@ -5006,7 +5013,7 @@ def raw_generate(
batch_count=batch_count batch_count=batch_count
) )
return GenerationResult( return GenerationResult(
out_batches=batch_encoded, prompt=prompt_tokens, is_whole_generation=False out_batches=batch_encoded, prompt=prompt_tokens, is_whole_generation=True
) )
elif model == "OAI": elif model == "OAI":
batch_encoded = oai_raw_generate( batch_encoded = oai_raw_generate(
@@ -5015,7 +5022,7 @@ def raw_generate(
batch_count=batch_count batch_count=batch_count
) )
return GenerationResult( return GenerationResult(
out_batches=batch_encoded, prompt=prompt_tokens, is_whole_generation=False out_batches=batch_encoded, prompt=prompt_tokens, is_whole_generation=True
) )
# Torch HF # Torch HF
@@ -5027,7 +5034,7 @@ def raw_generate(
batch_count=batch_count batch_count=batch_count
) )
return GenerationResult( return GenerationResult(
out_batches=batch_encoded, prompt=prompt_tokens, is_whole_generation=True out_batches=batch_encoded, prompt=prompt_tokens, is_whole_generation=False
) )
def tpu_raw_generate( def tpu_raw_generate(