[Bug]: bnb quant load error

theobjectivedad commented 2 weeks ago

Your current environment

Collecting environment information...
PyTorch version: 2.3.0+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A
OS: Ubuntu 22.04.4 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: version 3.29.3
Libc version: glibc-2.35
Python version: 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-6.5.0-35-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.1.105
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA RTX A6000
GPU 1: NVIDIA RTX A6000
GPU 2: NVIDIA RTX A6000
GPU 3: NVIDIA RTX A6000

Nvidia driver version: 535.171.04
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
CPU:
Architecture:                       x86_64
CPU op-mode(s):                     32-bit, 64-bit
Address sizes:                      48 bits physical, 48 bits virtual
Byte Order:                         Little Endian
CPU(s):                             64
On-line CPU(s) list:                0-63
Vendor ID:                          AuthenticAMD
Model name:                         AMD Ryzen Threadripper PRO 5975WX 32-Cores
CPU family:                         25
Model:                              8
Thread(s) per core:                 2
Core(s) per socket:                 32
Socket(s):                          1
Stepping:                           2
Frequency boost:                    enabled
CPU max MHz:                        7006.6401
CPU min MHz:                        1800.0000
BogoMIPS:                           7186.60
Flags:                              fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 invpcid_single hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin brs arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip pku ospke vaes vpclmulqdq rdpid overflow_recov succor smca fsrm
Virtualization:                     AMD-V
L1d cache:                          1 MiB (32 instances)
L1i cache:                          1 MiB (32 instances)
L2 cache:                           16 MiB (32 instances)
L3 cache:                           128 MiB (4 instances)
NUMA node(s):                       1
NUMA node0 CPU(s):                  0-63
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit:        Not affected
Vulnerability L1tf:                 Not affected
Vulnerability Mds:                  Not affected
Vulnerability Meltdown:             Not affected
Vulnerability Mmio stale data:      Not affected
Vulnerability Retbleed:             Not affected
Vulnerability Spec rstack overflow: Vulnerable: Safe RET, no microcode
Vulnerability Spec store bypass:    Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:           Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:           Mitigation; Retpolines; IBPB conditional; IBRS_FW; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                Not affected
Vulnerability Tsx async abort:      Not affected
Versions of relevant libraries:
[pip3] numpy==1.26.4
[pip3] torch==2.3.0
[pip3] triton==2.3.0
[conda] Could not collect ROCM Version: Could not collect
Aphrodite Version: 0.5.3
Aphrodite Build Flags:
CUDA Archs: 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX; ROCm: Disabled

Additionally:

transformers==4.40.0 bitsandbytes==0.43.1

🐛 Describe the bug

When loading CommandR+ (bnb/4bit) with the following command:

docker run -it -d --name=eleanor-aphrodite-main --restart=unless-stopped --shm-size=15g --ulimit memlock=-1 --ipc=host --entrypoint=python3 --gpus="device=0,1,2,3" --publish=7800:8000 --volume=/mnt/archive/models:/models:ro --health-cmd=timeout 5 bash -c 'cat < /dev/null > /dev/tcp/localhost/8000' --health-start-period=240s --health-interval=15s --health-timeout=8s --health-retries=3 --env=RAY_DEDUP_LOGS=1 --env=APHRODITE_ENGINE_ITERATION_TIMEOUT_S=120 quay.io/theobjectivedad/aphrodite:latest -m aphrodite.endpoints.openai.api_server --model /models/c4ai-command-r-plus-4bit --served-model-name c4ai-command-r-plus-4bit --quantization bnb --context-shift --tensor-parallel-size 4 --gpu-memory-utilization 0.90 --kv-cache-dtype auto --load-format safetensors --tokenizer-mode auto --dtype auto --response-role gpt --port 8000 --host 0.0.0.0

I get this error:

INFO: Tailing logs for eleanor-aphrodite-main. Press CTRL+C to stop.
Traceback (most recent call last):
  File "/usr/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/usr/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/app/aphrodite-engine/aphrodite/endpoints/openai/api_server.py", line 562, in <module>
    run_server(args)
  File "/app/aphrodite-engine/aphrodite/endpoints/openai/api_server.py", line 519, in run_server
    engine = AsyncAphrodite.from_engine_args(engine_args)
  File "/app/aphrodite-engine/aphrodite/engine/async_aphrodite.py", line 340, in from_engine_args
    engine_config = engine_args.create_engine_config()
  File "/app/aphrodite-engine/aphrodite/engine/args_tools.py", line 539, in create_engine_config
    model_config = ModelConfig(
  File "/app/aphrodite-engine/aphrodite/common/config.py", line 145, in __init__
    self._verify_quantization()
  File "/app/aphrodite-engine/aphrodite/common/config.py", line 215, in _verify_quantization
    raise ValueError(
ValueError: Quantization method specified in the model config (bitsandbytes) does not match the quantization method specified in the `quantization` argument (bnb).

The image was build from main on 6/11/2024.

As indicated in the error message, omitting --quantization bnb doesn't seem to have any effect, I the error is from reading the model's config.json file:

...
  "quantization_config": {
    "_load_in_4bit": true,
    "_load_in_8bit": false,
    "bnb_4bit_compute_dtype": "bfloat16",
    "bnb_4bit_quant_storage": "uint8",
    "bnb_4bit_quant_type": "fp4",
    "bnb_4bit_use_double_quant": true,
    "llm_int8_enable_fp32_cpu_offload": false,
    "llm_int8_has_fp16_weight": false,
    "llm_int8_skip_modules": null,
    "llm_int8_threshold": 6.0,
    "load_in_4bit": true,
    "load_in_8bit": false,
    "quant_method": "bitsandbytes"
  },
...

Some notes I took looking into this a little deeper...

The actual error is being triggered here, and is basically assuming quantization_config.quant_method in the model's config.json would be bnb ... I'm sure this changed at some point in transformers
In _verify_quantization() , we may want to keep bnb and add bitsandbytes to maintain compatibility and reduce confusion.
Additionally, there are some default values in the same method that get set when loading a model in 4/8 bit mode via bnb
The transformers config doc is here, I'm still searching to see exactly when this change was made (not that it matters much)

Lastly, I'm hesitant to fly solo and just submit a PR on this one since the current bnb quant config is pretty tightly integrated into aphrodite/quantization/bitsandbytes.py, not sure on the best way to proceed.

AlpinDale commented 2 weeks ago

The quant name in aphrodite is unfortunately a bit misleading - I intend to fix this with the next release. The load_in_4bit quant isn't actually bitsandbytes, it's SmoothQuant+. We don't allow loading bnb weights directly yet. This will also be addressed with the next release.

Note that SQ+ is faster and offers better quality than bnb 4bit. bnb reduces throughput compared to fp16, while sq+ increases it by close to 3x.

theobjectivedad commented 2 weeks ago

Got it, ty for looking at this and helping me understand. Do you want me to close this issue?

AlpinDale commented 2 weeks ago

Might be a good idea to keep it open in case someone else has the same issue. I'll close it myself once we have real bitsandbytes support.

PygmalionAI / aphrodite-engine

[Bug]: bnb quant load error #513

Your current environment

🐛 Describe the bug