File size: 5,180 Bytes
956a1c1 5982e82 956a1c1 9d29a15 cd33bcc 9d29a15 cd33bcc 956a1c1 3fffd5b 956a1c1 3fffd5b 0da14e4 956a1c1 cd33bcc 956a1c1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 | {
"release_name": "Ling-3.0-Flash-CIRU-int4-Strix-native",
"release_archive": "Ling-3.0-Flash-CIRU-int4-Strix-native.tar.gz",
"release_type": "runtime-distribution",
"weights_changed": false,
"base_model": {
"repo_id": "inclusionAI/Ling-3.0-flash-int4",
"revision": "ca3ea63b0255d212c4fe6020db9e0a51ce136006",
"format": "compressed-tensors packed INT4 safetensors",
"shards": 24,
"shard_bytes": 77012299464,
"native_context": 262144
},
"recommended_live_generation": {
"source": "https://huggingface.co/inclusionAI/Ling-3.0-flash-int4",
"source_revision": "ca3ea63b0255d212c4fe6020db9e0a51ce136006",
"temperature": 0.6,
"top_p": 0.95,
"top_k": 20,
"enable_thinking": true
},
"runtime": {
"upstream_repo": "https://github.com/vllm-project/vllm.git",
"upstream_base": "d35eb6c44071ea806018841c490f0d2f3219c485",
"ciru_head": "838616875c5dd4913d9f753221bf05f64cb7a7ed",
"commits_ahead": 17,
"net_files_changed": 10,
"net_insertions": 477,
"net_deletions": 28,
"engine_revision": "2026-08-10-startup-and-installer-hardening",
"rocr_idle_fix": {
"upstream_repo": "https://github.com/ROCm/rocm-systems.git",
"upstream_commit": "44be71b52284948e58c93f65f46910399773fdcd",
"library": "libhsa-runtime64.so.1.21.0",
"source_changes": "none; host-native GCC rebuild of the pinned wheel source",
"activation": "checksum-verified LD_PRELOAD in packaged launchers",
"rollback_environment": "CIRU_DISABLE_ROCR_IDLE_FIX=1"
}
},
"validated_stack": {
"python": "3.12.12",
"torch": "2.13.0+rocm7.15.0a20260728",
"hip": "7.15.0",
"triton": "3.8.0+git4cff872c.rocm7.15.0a20260728",
"flash_attention": "2.8.3",
"gpu": "AMD Radeon 8060S",
"architecture": "gfx1151"
},
"profiles": {
"native_256k": {"max_model_len": 262144, "max_num_seqs": 6, "max_num_batched_tokens": 8192, "gpu_memory_utilization": 0.72, "kv_pool_tokens_observed": 1457313, "full_context_concurrency_observed": 5.56},
"experimental_yarn_1m": {"max_model_len": 1048576, "max_num_seqs": 2, "max_num_batched_tokens": 8192, "gpu_memory_utilization": 0.82, "native_training": false}
},
"platforms": {
"validated": ["NixOS native Linux on gfx1151"],
"documented_unvalidated": ["Ubuntu/Debian native Linux", "Fedora native Linux", "Arch native Linux", "Windows 11 WSL2 with AMD ROCDXG"],
"unsupported": ["native Windows vLLM"]
},
"performance": {
"host_scope": "single AMD Strix Halo gfx1151 host",
"public_starting_target_only_tokens_per_second": 0.268944,
"release_target_only_tokens_per_second": 21.4407,
"release_native_mtp_k1_effective_tokens_per_second": 26.2343,
"release_target_only_multiple": 79.72,
"release_native_mtp_k1_multiple": 97.55,
"comparison_scope": "full campaign progression; starting TG8 and final TG32, not a single-variable matched A/B",
"exact_2k_k1_concurrency": {
"protocol": "exact 2048 input tokens, forced 256 output tokens, K1 MTP, unique cache salts, no prefix reuse, max_num_batched_tokens=8192",
"repetitions_per_concurrency": 3,
"all_18_runs_valid": true,
"no_speed_regression": true,
"c1": {"pp_tokens_per_second": 523.62, "aggregate_decode_tokens_per_second": 26.79, "wall_seconds": 13.45},
"c2": {"pp_tokens_per_second": 520.98, "aggregate_decode_tokens_per_second": 38.23, "wall_seconds": 20.86},
"c3": {"pp_tokens_per_second": 514.93, "aggregate_decode_tokens_per_second": 46.17, "wall_seconds": 28.08},
"c4": {"pp_tokens_per_second": 523.95, "aggregate_decode_tokens_per_second": 57.12, "wall_seconds": 32.91},
"c5": {"pp_tokens_per_second": 510.72, "aggregate_decode_tokens_per_second": 60.55, "wall_seconds": 39.67},
"c6": {"pp_tokens_per_second": 520.49, "aggregate_decode_tokens_per_second": 63.51, "wall_seconds": 42.60},
"validated_max_concurrency": 6
},
"rocm_mla_lse_fix_validation": {
"runtime_commit": "91fab4e6dcaa683911655630dae8a5b140a7cb91",
"focused_tests_passed": 31,
"cuda_only_tests_deselected": 1,
"reproducer_concurrency": 5,
"reproducer_scenarios_passed": 5,
"reproducer_request_error_rate": 0.0,
"server_process_survived": true,
"result_sha256": "f819148f8d49342a73fac2c8be6ce6cd08dfca01c238e94dd586bbd6e5894cd7"
},
"rocr_idle_fix_validation": {
"baseline_cpu_temperature_c": 98.0,
"corrected_idle_temperature_c_range": [38.8, 54.2],
"hottest_idle_runtime_thread_cpu_percent_max": 1.4,
"response_tokens": 256,
"response_seconds": 9.731,
"response_tokens_per_second": 26.307,
"exact_api_output_passed": true
},
"retained_60k_prompt_512_output": {
"native_source": "fabbc7e928dfa64c29dcdf0f795b05246555f4ae",
"native_pp_tokens_per_second": 419.334843,
"native_tg_tokens_per_second": 19.924752,
"atomic_chat_pp_tokens_per_second": 265.63,
"atomic_chat_tg_tokens_per_second": 25.11,
"rocmfp4_pp_tokens_per_second": 249.23,
"rocmfp4_tg_tokens_per_second": 19.00,
"refresh_required_for_release_head": true
}
}
}
|