Commit 627973d
committed
Massive updates across ColossalAI applications
Broad refactor and feature updates across the ColossalAI repository: updates to Colossal-LLaMA (dataset, tokenizer, model init, utils, examples, training scripts), ColossalChat (coati modules, quantization, ray integration, trainers, benchmarks, examples, tests), ColossalEval (datasets, evaluators, metrics, examples), ColossalMoE (models, utils, train/infer), ColossalQA (retrieval, loaders, text splitter, webui examples), and many core subsystems (accelerator, analyzer, auto_parallel, autochunk, booster, tensor_shard node handlers and strategy generators). Also includes CI/workflow script changes and assorted utility/test improvements. These changes consolidate bug fixes, feature enhancements, new examples and tests to improve functionality and reliability across the codebase.1 parent a115c7f commit 627973d
1,615 files changed
Lines changed: 1612 additions & 338 deletions
File tree
- Src
- Main_Scripts
- ColossalAI
- .github/workflows/scripts
- example_checks
- applications
- Colossal-LLaMA
- colossal_llama
- dataset
- model
- tokenizer
- utils
- ColossalChat
- benchmarks
- ray
- coati
- dataset
- experience_buffer
- experience_maker
- models
- quant
- llama_gptq
- ray
- callbacks
- trainer
- callbacks
- utils
- examples
- community
- peft
- ray
- data_preparation_scripts
- inference
- web_chatbot
- ray
- training_scripts
- tests
- ColossalEval
- colossal_eval
- dataset
- evaluate
- dataset_evaluator
- models
- utils
- examples
- dataset_evaluation
- gpt_evaluation
- ColossalMoE
- colossal_moe
- models
- tests
- ColossalQA
- colossalqa
- chain
- memory
- retrieval_qa
- data_loader
- local
- prompt
- text_splitter
- examples
- webui_demo
- tests
- colossalai
- _C
- _analyzer
- _subclasses
- fx
- passes
- tracer
- accelerator
- amp
- naive_amp
- grad_scaler
- mixed_precision_mixin
- auto_parallel
- checkpoint
- meta_profiler
- meta_registry
- offload
- passes
- pipeline_shard
- tensor_shard
- node_handler
- strategy
- solver
- utils
- autochunk
- booster
- mixed_precision
- plugin
- checkpoint_io
- cli
- check
- launcher
- cluster
- context
- device
- fx
- codegen
- passes
- experimental
- profiler
- experimental
- profiler_function
- profiler_module
- tracer
- bias_addition_patch
- patched_bias_addition_function
- patched_bias_addition_module
- meta_patch
- patched_function
- patched_module
- inference
- engine
- modeling
- policies
- kv_cache
- quant
- gptq
- cai_gptq
- smoothquant
- models
- interface
- kernel
- jit
- triton
- lazy
- legacy
- amp
- apex_amp
- naive_amp
- torch_amp
- builder
- communication
- context
- process_group_initializer
- random
- engine
- gradient_accumulation
- gradient_handler
- schedule
- inference
- dynamic_batching
- hybridengine
- modeling
- polices
- pipeline
- benchmark
- quant
- gptq
- cai_gptq
- smoothquant
- models
- serving
- ray_serve
- torch_serve
- tensor_parallel
- modeling
- policies
- nn
- _ops
- layer
- colossalai_layer
- parallel_1d
- parallel_2d
- parallel_2p5d
- parallel_3d
- parallel_sequence
- utils
- vanilla
- wrapper
- loss
- metric
- parallel
- layers
- cache_embedding
- pipeline
- middleware
- adaptor
- rpc
- registry
- tensor
- trainer
- hooks
- utils
- checkpoint
- data_sampler
- profiler
- legacy
- zero
- gemini
- ophooks
- paramhooks
- init_ctx
- shard_utils
- sharded_model
- sharded_optim
- sharded_param
- logging
- moe
- nn
- layer
- loss
- lr_scheduler
- optimizer
- pipeline
- schedule
- shardformer
- examples
- layer
- modeling
- chatglm2_6b
- policies
- shard
- tensor
- d_tensor
- moe_tensor
- padded_tensor
- testing
- utils
- model
- multi_tensor_apply
- rank_recorder
- tensor_detector
- zero
- gemini
- chunk
- memory_tracer
- low_level
- bookkeeping
- examples
- community
- fp8/mnist
- roberta
- preprocessing
- pretraining
- model
- utils
- images
- diffusion
- ldm
- data
- models
- diffusion
- dpm_solver
- modules
- diffusionmodules
- distributions
- encoders
- image_degradation
- midas
- midas
- scripts
- tests
- dreambooth
- resnet
- vit
- inference
- language
- bert
- commons
- gpt
- experiments
- auto_offload
- auto_parallel
- pipeline_parallel
- gemini
- commons
- hybridparallelism
- titans
- configs
- dataset
- model
- grok-1
- llama
- openmoe
- benchmark
- model
- opt
- palm
- palm_pytorch
- tutorial
- auto_parallel
- hybrid_parallel
- large_batch_optimizer
- new_api
- cifar_resnet
- cifar_vit
- glue_bert
- opt
- inference
- benchmark
- script
- process-opt-175b
- opt
- sequence_parallel
- data
- datasets
- test
- tokenizer
- loss_func
- lr_scheduler
- model
- layers
- extensions
- cpu_adam
- csrc
- cuda
- flash_attention
- layernorm
- moe
- optimizer
- softmax
- tests
- kit
- model_zoo
- custom
- diffusers
- timm
- torchaudio
- torchrec
- torchvision
- transformers
- test_analyzer
- test_fx
- test_subclasses
- test_auto_parallel
- test_ckpt_solvers
- test_offload
- test_pass
- test_tensor_shard
- test_gpt
- test_metainfo
- test_node_handler
- test_autochunk
- test_autochunk_alphafold
- test_autochunk_diffuser
- test_autochunk_transformer
- test_autochunk_vit
- test_booster
- test_mixed_precision
- test_plugin
- test_checkpoint_io
- test_cluster
- test_config
- test_device
- test_fx
- test_codegen
- test_meta
- test_pipeline
- test_hf_model
- test_timm_model
- test_topo
- test_torchvision
- test_profiler
- test_tracer
- test_hf_model
- test_timm_model
- test_torchaudio_model
- test_torchrec_model
- test_torchvision_model
- test_gptq
- test_infer
- test_lazy
- test_legacy
- test_amp
- test_comm
- test_context
- configs
- test_data
- test_engine
- test_layers
- test_1d
- checks_1d
- test_2d
- checks_2d
- test_2p5d
- checks_2p5d
- test_3d
- checks_3d
- test_sequence
- checks_seq
- test_pipeline
- test_tensor
- common_utils
- core
- test_trainer
- test_pipeline
- test_utils
- test_checkpoint
- test_zero
- test_moe
- test_optimizer
- test_pipeline
- test_pipeline_utils
- test_schedule
- test_shardformer
- test_hybrid_parallel_grad_clip_norm
- test_layer
- test_model
- test_smoothquant
- test_tensor
- test_dtensor
- test_zero
- test_gemini
- test_low_level
- backend
- config
- core
- moe_inference_runtime
- backends
- export
- monitoring
- security
- training
- utils
- tests
Some content is hidden
Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.
Binary file not shown.
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
| 1 | + | |
1 | 2 | | |
2 | 3 | | |
3 | 4 | | |
| |||
Lines changed: 1 addition & 0 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
| 1 | + | |
1 | 2 | | |
2 | 3 | | |
3 | 4 | | |
| |||
Lines changed: 1 addition & 0 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
| 1 | + | |
1 | 2 | | |
2 | 3 | | |
3 | 4 | | |
| |||
Lines changed: 1 addition & 0 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
| 1 | + | |
1 | 2 | | |
2 | 3 | | |
3 | 4 | | |
| |||
Lines changed: 1 addition & 0 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
| 1 | + | |
1 | 2 | | |
2 | 3 | | |
3 | 4 | | |
| |||
Lines changed: 1 addition & 0 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
| 1 | + | |
1 | 2 | | |
2 | 3 | | |
3 | 4 | | |
| |||
Lines changed: 1 addition & 0 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
| 1 | + | |
1 | 2 | | |
2 | 3 | | |
3 | 4 | | |
| |||
Lines changed: 1 addition & 0 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
| 1 | + | |
1 | 2 | | |
2 | 3 | | |
3 | 4 | | |
| |||
Lines changed: 1 addition & 0 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
| 1 | + | |
1 | 2 | | |
2 | 3 | | |
3 | 4 | | |
| |||
0 commit comments