GitLab

Co-authored-by: Varun Sundar Rabindranath <varunsundar08@gmail.com>
Co-authored-by: Varun Sundar Rabindranath <varun@neuralmagic.com>

Co-authored-by: Cody Yu <hao.yu.cody@gmail.com>

Co-authored-by: Varun Sundar Rabindranath <varun@neuralmagic.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-neuralmagic@users.noreply.github.com>

Revert "[Kernel] Marlin_24: Ensure the mma.sp instruction is using the ::ordered_metadata modifier (introduced with PTX 8.5)" (#5149)

[Kernel] Marlin_24: Ensure the mma.sp instruction is using the ::ordered_metadata modifier (introduced with PTX 8.5) (#5136)

Co-authored-by: Varun Sundar Rabindranath <varunsundar08@gmail.com>
Co-authored-by: Varun Sundar Rabindranath <varun@neuralmagic.com>

Pass the CUDA stream into the CUTLASS GEMMs, to avoid future issues with CUDA graphs

Co-authored-by: Robert Shaw <rshaw@neuralmagic.com>

Previously FP8 static scaling works if the scales are overestimating the maxima of all activation tensors during computation. However this will not always be the case even if the scales were calibrated very carefully. For example, with the activations in my checkpoint

https://huggingface.co/pcmoritz/Mixtral-8x7B-v0.1-fp8-act-scale

(which was calibrated on https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k), I'm getting the following mostly random performance on MMLU:

|      Groups      |Version|Filter|n-shot|Metric|Value |   |Stderr|
|------------------|-------|------|-----:|------|-----:|---|-----:|
|mmlu              |N/A    |none  |     0|acc   |0.2295|±  |0.0035|
| - humanities     |N/A    |none  |     5|acc   |0.2421|±  |0.0062|
| - other          |N/A    |none  |     5|acc   |0.2398|±  |0.0076|
| - social_sciences|N/A    |none  |     5|acc   |0.2171|±  |0.0074|
| - stem           |N/A    |none  |     5|acc   |0.2125|±  |0.0073|
With the fix in this PR where the scaled activations are clamped between [-std::numeric_limits<c10::Float8_e4m3fn>::max(), std::numeric_limits<c10::Float8_e4m3fn>::max()] to make sure there are no NaNs, the performance is

|      Groups      |Version|Filter|n-shot|Metric|Value |   |Stderr|
|------------------|-------|------|-----:|------|-----:|---|-----:|
|mmlu              |N/A    |none  |     0|acc   |0.7008|±  |0.0036|
| - humanities     |N/A    |none  |     5|acc   |0.6453|±  |0.0065|
| - other          |N/A    |none  |     5|acc   |0.7692|±  |0.0072|
| - social_sciences|N/A    |none  |     5|acc   |0.8083|±  |0.0070|
| - stem           |N/A    |none  |     5|acc   |0.6115|±  |0.0083|
This is not perfect yet but is getting very close to the FP16 / dynamic activation scale performance.

Co-authored-by: alexm <alexm@neuralmagic.com>
Co-authored-by: mgoin <michael@neuralmagic.com>

Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>

This PR addresses the Marlin kernel H100 crash that was reported here: neuralmagic#187.
The reason for the crash was the inline PTX assembly that introduced the async_copy with streaming behavior. The solution is to use the more standard PTX for async_copy (without the fractional L2 policy for "evict_first"). There is no performance difference between standard async_copy PTX and the previous one.

This PR is the first step towards fixing https://github.com/vllm-project/vllm/pull/3208

It implements dynamic per-tensor scaling (see https://github.com/vllm-project/vllm/pull/4118), so users do not need to compute activation scales on a calibration dataset and they also don't need to convert their model checkpoints. It is enough to specify the `quantization="fp8"` argument. You can try out the PR like this:

```python
from vllm import LLM, SamplingParams

prompts = [
    "Hello, my name is",
    "The president of the United States is",
    "The capital of France is",
    "The future of AI is",
]
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)

llm = LLM(model="mistralai/Mixtral-8x7B-Instruct-v0.1", tensor_parallel_size=2, quantization="fp8")

outputs = llm.generate(prompts, sampling_params)

# Print the outputs.
for output in outputs:
    prompt = output.prompt
    generated_text = output.outputs[0].text
    print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```

**Performance**: For this PR, the focus is on making the code clean (while still trying to get reasonable performance), there is a bunch of optimizations that we will submit as a follow up PR that significantly improve the performance (similar to the numbers in https://github.com/vllm-project/vllm/pull/3954). With this PR, the results are as follows:

<img width="725" alt="Screenshot 2024-04-21 at 1 31 50 PM" src="https://github.com/vllm-project/vllm/assets/113316/d8fe1118-07a0-4d4e-8530-37a77d465a03">


**Accuracy**: The accuracy with this PR on MMLU on `mistralai/Mixtral-8x7B-v0.1` is as follows:

```
|      Groups      |Version|Filter|n-shot|Metric|Value |   |Stderr|
|------------------|-------|------|-----:|------|-----:|---|-----:|
|mmlu              |N/A    |none  |     0|acc   |0.7018|±  |0.0036|
| - humanities     |N/A    |none  |     5|acc   |0.6472|±  |0.0065|
| - other          |N/A    |none  |     5|acc   |0.7673|±  |0.0072|
| - social_sciences|N/A    |none  |     5|acc   |0.8099|±  |0.0070|
| - stem           |N/A    |none  |     5|acc   |0.6131|±  |0.0083|
```
this compares favorably with the fp16 results which are
```
|      Groups      |Version|Filter|n-shot|Metric|Value |   |Stderr|
|------------------|-------|------|-----:|------|-----:|---|-----:|
|mmlu              |N/A    |none  |     0|acc   |0.7020|±  |0.1313|
| - humanities     |N/A    |none  |     5|acc   |0.6425|±  |0.1349|
| - other          |N/A    |none  |     5|acc   |0.7744|±  |0.1038|
| - social_sciences|N/A    |none  |     5|acc   |0.8131|±  |0.0695|
| - stem           |N/A    |none  |     5|acc   |0.6108|±  |0.1383|
```

Happy hacking!

Co-authored-by: mgoin <michael@neuralmagic.com>

Co-authored-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>
Co-authored-by: HaiShaw <hixiao@gmail.com>
Co-authored-by: AdrianAbeyta <Adrian.Abeyta@amd.com>
Co-authored-by: Matthew Wong <Matthew.Wong2@amd.com>
Co-authored-by: root <root@gt-pla-u18-08.pla.dcgpu>
Co-authored-by: mawong-amd <156021403+mawong-amd@users.noreply.github.com>
Co-authored-by: ttbachyinsda <ttbachyinsda@outlook.com>
Co-authored-by: guofangze <guofangze@kuaishou.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: jacobthebanana <50071502+jacobthebanana@users.noreply.github.com>
Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>

Co-authored-by: Robert Shaw <114415538+rib-2@users.noreply.github.com>
Co-authored-by: alexm <alexm@neuralmagic.com>

Co-authored-by: Chunan Zeng <chunanzeng@Chunans-Air.attlocal.net>

Co-authored-by: zhaoyang <zhao.yang16@zte.com.cn>
Co-authored-by: Zhuohan Li <zhuohan123@gmail.com>