fix(model): Fix reasoning output bug

This commit is contained in:
Fangyin Cheng
2025-03-04 17:51:13 +08:00
parent e3a25de7f7
commit c8e252c4de
27 changed files with 1332 additions and 2650 deletions

View File

@@ -0,0 +1,4 @@
# DB-GPT Accelerator Module
Building across multiple platforms and hardware is complex, and the DB-GPT Accelerator aims to provide compatibility handling for this, offering as consistent an interface as possible for other core models.

View File

@@ -1,5 +1,5 @@
[project]
name = "dbgpt-accelerator"
name = "dbgpt-acc-auto"
version = "0.7.0"
description = "Add your description here"
authors = [
@@ -16,21 +16,6 @@ Documentation = "http://docs.dbgpt.cn/docs/overview"
Repository = "https://github.com/eosphoros-ai/DB-GPT.git"
Issues = "https://github.com/eosphoros-ai/DB-GPT/issues"
[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"
[tool.hatch.build.targets.wheel]
packages = ["src/dbgpt_accelerator"]
exclude = [
"src/dbgpt_accelerator/**/tests",
"src/dbgpt_accelerator/**/tests/*",
"src/dbgpt_accelerator/tests",
"src/dbgpt_accelerator/tests/*",
"src/dbgpt_accelerator/**/examples",
"src/dbgpt_accelerator/**/examples/*"
]
[project.optional-dependencies]
# Auto install dependencies
auto = [
@@ -76,10 +61,10 @@ vllm = [
# Just support GPU version on Linux
"vllm>=0.7.0; sys_platform == 'linux'",
]
#vllm_pascal = [
# vllm_pascal = [
# # https://github.com/sasha0552/pascal-pkgs-ci
# "vllm-pascal==0.7.2; sys_platform == 'linux'"
#]
# ]
quant_bnb = [
"bitsandbytes>=0.39.0; sys_platform == 'win32' or sys_platform == 'linux'",
"accelerate"
@@ -103,6 +88,10 @@ quant_gptq = [
"optimum",
"auto-gptq",
]
flash_attn = [
# "torch>=2.2.1",
"dbgpt-acc-flash-attn"
]
[dependency-groups]
auto = [

View File

@@ -0,0 +1,3 @@
# DB-GPT-Accelerator for Flash Attention
Wrapper for the Flash Attention module in the DB-GPT-Accelerator.

View File

@@ -0,0 +1,24 @@
# Install the flash-attn package for uv
# https://github.com/astral-sh/uv/issues/2252#issuecomment-2624150395
[project]
name = "dbgpt-acc-flash-attn"
version = "0.1.0"
description = "Add your description here"
readme = "README.md"
requires-python = ">=3.10"
dependencies = []
[dependency-groups]
build = [
"setuptools>=75.8.0",
]
direct = [
"torch>=2.2.1",
]
main = [
"flash-attn>=2.5.8",
]
[tool.uv]
default-groups = ["build", "direct", "main"]
no-build-isolation-package = ["flash-attn"]