model string | device string | bits int64 | sources dict | results list |
|---|---|---|---|---|
mistralai/Mistral-7B-v0.1 | AMD Instinct MI300X VF | 3 | {
"ratio": "/root/workspace/amd-experiments/results/bench_runtime_ratio_all_methods.json",
"speed": "/root/workspace/amd-experiments/results/bench_compress_decompress_recheck.json",
"quality": "/root/workspace/amd-experiments/results/bench_ppl_all_methods_quality_recheck.json"
} | [
{
"method": "turbo",
"bits": 3,
"seq_len": 2048,
"batch_size": 1,
"cache_layers": 32,
"cache_kv_heads": 8,
"cache_head_dim": 128,
"kv_bytes_fp16": 268435456,
"kv_bytes_compressed_materialized": 54525952,
"ratio_calculated_layout": 4.923077,
"ratio_observed_runtime": 4.923... |
amd-experiments: MI300X KV-cache compression benchmark results
Benchmark result files for amd-experiments: a four-way comparison of TurboQuant, IsoQuant, PlanarQuant and RotorQuant KV-cache compression on an AMD Instinct MI300X (gfx942) with Triton/ROCm kernels and Mistral-7B-v0.1 shapes.
No model weights are released. The GitHub repo contains the kernels and benchmark scripts (and a README, video and report); this dataset repo holds the raw bench_*.json result files and the derived summary readme_data.json only.
Headline (from results/readme_data.json)
| method | compression | compress GB/s | prefill K tok/s at 32K | cosine (3-bit) |
|---|---|---|---|---|
| TurboQuant (TQ3) | 4.92x | 2.9 | 42 | 0.983 |
| IsoQuant | 4.92x | 21.8 | 892 | 0.983 |
| PlanarQuant | 4.92x | 18.7 | 1,126 | 0.983 |
| RotorQuant | 4.92x | 17.3 | 855 | 0.983 |
- A fused Triton TQ3 attention kernel is 0.28x of fp16 SDPA speed at 1K tokens and about 1.9x at 32K (B=1).
- In vLLM serving of Mistral-7B (1024 in / 256 out, 32 prompts) output throughput is flat across fp16, TurboQuant decompress and TurboQuant fused (2,425 to 2,435 tok/s).
- Single GPU model, one LLM family; the needle-in-a-haystack test is synthetic.
Files
results/bench_*.json: one file per benchmark (compress/decompress bandwidth, prefill, perplexity and cosine quality, NIAH, fused-attention timings, vLLM A/B). Load with json.load.
Scripts to regenerate them: https://github.com/Bobliuuu/amd-experiments (benchmarks/, run_all_benchmarks.sh).
- Downloads last month
- 61