whisper-blaze 0.1.8

Hopper-native CUDA kernels for Whisper large-v3 on H100 GPU

whisper-blaze

Hopper-native CUDA kernels for Whisper large-v3 on NVIDIA H100 GPUs.

Replaces standard PyTorch operations with hand-tuned CUDA kernels that exploit H100-specific hardware:

• WGMMA (Warpgroup MMA) GEMM in FP16 and FP8 (E4M3 / E5M2)
• TMA (Tensor Memory Accelerator) async bulk copy
• Flash Attention 3 for encoder self-attention, decoder self/cross-attention
• Fused residual + LayerNorm / RMSNorm
• GPU mel spectrogram (replaces CPU librosa/HuggingFace preprocessor)
• FP8 quantize/dequantize with per-tensor scaling
• Dynamic cross-request batching — fuses concurrent API calls into one model.generate() pass to fill all 80 GB of H100 VRAM

Requirements

Component	Version
GPU	NVIDIA H100 (Hopper, SM90)
CUDA toolkit	12.2+ (12.6 recommended)
PyTorch	2.1.0+ with matching CUDA
Python	3.9+
OS	Linux x86_64

Installation

Step 1 — Install PyTorch with CUDA support (if you haven't already):

pip install torch --index-url https://download.pytorch.org/whl/cu124

Step 2 — Install whisper-blaze:

pip install whisper-blaze --no-build-isolation

--no-build-isolation is required — it tells pip to use your existing PyTorch instead of fetching it into an isolated build environment.

From source:

git clone https://github.com/YOUR_USERNAME/whisper-blaze.git
cd whisper-blaze
pip install -e . --no-build-isolation

If your CUDA toolkit isn't at /usr/local/cuda, set CUDA_HOME first:

export CUDA_HOME=/usr/local/cuda-12.6

Quick Start

from whisper_blaze import WhisperBlaze
from whisper_blaze.precision import mixed_fp8

model = WhisperBlaze.from_pretrained(
    "openai/whisper-large-v3",
    precision=mixed_fp8(),
)

# Single file — numpy array or torch tensor, float32, 16 kHz
# 1D [samples] or 2D [channels, samples] both accepted
result = model.transcribe(audio, language="en")
print(result["text"])

Batch Transcription

transcribe_batch() accepts multiple audio files and fuses all their 30-second chunks into a single model.generate() call, maximising VRAM utilisation on an 80 GB H100.

# results is a list of dicts, one per input audio
results = model.transcribe_batch(
    [audio1, audio2, audio3],
    language="en",
    task="transcribe",
)
for r in results:
    print(r["text"])

Why it matters: a single 15-minute file uses ~40 GB VRAM. With transcribe_batch() you can process a second 15-minute file in the same GPU pass, using ~78 GB — the remaining 40 GB that would otherwise sit idle.

VRAM usage guide

Batch contents	Approx. VRAM	Notes
1 × 15-min audio	~40 GB	baseline
2 × 15-min audio	~78 GB	safe on 80 GB H100
4 × 5-min audio	~45 GB	short calls batch very efficiently
8 × 2-min audio	~40 GB	high-concurrency call-centre workload

Rule of thumb: ~1.15 GB per 30-second chunk plus ~6 GB for model weights.

Precision Presets

Preset	When to use
full_fp16()	Maximum quality, no quantization
mixed_fp8()	Recommended — FP8 on FFN/QKV, FP16 on attention
aggressive_fp8()	Maximum throughput, FP8 everywhere

from whisper_blaze.precision import full_fp16, mixed_fp8, aggressive_fp8

model = WhisperBlaze.from_pretrained(precision=aggressive_fp8())

Serving at Scale

For production deployments, pair whisper-blaze with a dynamic batching API server that keeps a pool of concurrent requests in-flight and automatically groups them into GPU batches:

Client pool (10 concurrent)
        │
        ▼
  FastAPI server              ← collect requests for 400 ms
        │
        ▼
  transcribe_batch()          ← one model.generate() for the whole batch
        │
        ▼
  Results returned individually

Throughput on a single H100 (80 GB) with 10-minute audio:

Mode	Calls/day
Single request, serialised	~30,000
Dynamic batching (2 concurrent)	~60,000
Dynamic batching (mixed lengths)	up to ~90,000

GPU Mel Spectrogram

from whisper_blaze import WhisperBlazeProcessor

proc = WhisperBlazeProcessor(device="cuda")
mel = proc(audio_tensor, sampling_rate=16000)   # [1, 128, T] fp16 on GPU

# Long audio with overlapping chunks
mels = proc.process_chunks(long_audio, sampling_rate=16000, overlap_s=1.0)

Direct Kernel API

import torch
import whisper_blaze_kernels as k

# FP8 quantize / dequantize
x = torch.randn(512, 512, dtype=torch.float16, device="cuda")
fp8, scale = k.quantise_e4m3(x)
x_back = k.dequantise_e4m3(fp8, scale, [512, 512])

# Fused residual + LayerNorm
out = k.layernorm_fused(hidden, residual, gamma, beta, 1e-5)

# Fused RMSNorm
out = k.rmsnorm_fused(hidden, residual, gamma, 1e-5)

# Flash Attention 3
out = k.encoder_self_attn(Q, K, V)    # no causal mask
out = k.decoder_self_attn(Q, K, V)    # causal mask
out = k.decoder_cross_attn(Q, K, V)   # no causal mask

# GPU mel spectrogram
mel = k.mel_spectrogram(audio_cpu_float32)  # → [1, 128, T] fp16 on GPU

Troubleshooting

RuntimeError: CUDA version mismatch — Your PyTorch was compiled against a different CUDA version than your system toolkit. Reinstall PyTorch from the correct index:

pip install torch --index-url https://download.pytorch.org/whl/cu124

ninja not found — Install ninja for faster builds:

pip install ninja

nvcc does not support sm_90a — Upgrade your CUDA toolkit to 12.2+. The H100 Hopper architecture requires sm_90a.

License

MIT