Download results/bench_measured_cache_memory.json from Bobliu/amd-experiments: direct link, hf CLI and curl.
- Browser
- Download file 1.64 kB
-
https://huggingface.co/datasets/Bobliu/amd-experiments/resolve/main/results/bench_measured_cache_memory.json
- Command line
-
hf download hf://datasets/Bobliu/amd-experiments/results/bench_measured_cache_memory.json
-
curl -L -o bench_measured_cache_memory.json https://huggingface.co/datasets/Bobliu/amd-experiments/resolve/main/results/bench_measured_cache_memory.json
1.64 kB
| { | |
| "model": "mistralai/Mistral-7B-v0.1", | |
| "device": "AMD Instinct MI300X VF", | |
| "seq_len": 512, | |
| "batch_size": 1, | |
| "bits": 3, | |
| "results": [ | |
| { | |
| "method": "turbo", | |
| "bits": 3, | |
| "label": "turbo3", | |
| "seq_len": 512, | |
| "batch_size": 1, | |
| "cache_layers": 32, | |
| "cache_kv_heads": 8, | |
| "cache_head_dim": 128, | |
| "compress_ms": 76.508, | |
| "kv_bytes_fp16": 67108864, | |
| "kv_bytes_compressed_measured": 13631488, | |
| "compression_ratio_measured": 4.923076923076923 | |
| }, | |
| { | |
| "method": "planar", | |
| "bits": 3, | |
| "label": "planar3", | |
| "seq_len": 512, | |
| "batch_size": 1, | |
| "cache_layers": 32, | |
| "cache_kv_heads": 8, | |
| "cache_head_dim": 128, | |
| "compress_ms": 412.916, | |
| "kv_bytes_fp16": 67108864, | |
| "kv_bytes_compressed_measured": 34603008, | |
| "compression_ratio_measured": 1.9393939393939394 | |
| }, | |
| { | |
| "method": "iso", | |
| "bits": 3, | |
| "label": "iso3", | |
| "seq_len": 512, | |
| "batch_size": 1, | |
| "cache_layers": 32, | |
| "cache_kv_heads": 8, | |
| "cache_head_dim": 128, | |
| "compress_ms": 11.278, | |
| "kv_bytes_fp16": 67108864, | |
| "kv_bytes_compressed_measured": 34603008, | |
| "compression_ratio_measured": 1.9393939393939394 | |
| }, | |
| { | |
| "method": "rotor", | |
| "bits": 3, | |
| "label": "rotor3", | |
| "seq_len": 512, | |
| "batch_size": 1, | |
| "cache_layers": 32, | |
| "cache_kv_heads": 8, | |
| "cache_head_dim": 128, | |
| "compress_ms": 11.512, | |
| "kv_bytes_fp16": 67108864, | |
| "kv_bytes_compressed_measured": 34865152, | |
| "compression_ratio_measured": 1.9248120300751879 | |
| } | |
| ] | |
| } |