ginfer
catalog
ginfer-catalog / catalog.json
SectileRon's picture
Flash 2-bit: name the package pure 2-bit
cb85b73 verified
Raw History Blame Contribute Delete
312 kB
{ "manifest_version": 1, "models": [ { "description": "GInfer packages of Muse-Glimmer-30B in INT. They run only on GInfer and GChat.", "developer": "SectileLabs", "downloads": 0, "library_name": "ginfer", "model_name": "SectileLabs/muse_glimmer_30b_int", "name": "Muse-Glimmer-30B INT", "releases": [ { "bytes": 19776776704, "capabilities": [ "tools", "vision" ], "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "launch_profiles": [ { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c1-128k-int8", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 1 request \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 458227712, "full_context_requests": 1, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c1-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C1 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 899682240 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c1-131072-windows-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 131,072 context \u00b7 C1 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 899682240 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c1-131072-windows-calculated-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C1 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 899682240 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c2-128k-int8", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 2 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 458227712, "full_context_requests": 2, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c2-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C2 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 1799364544 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c2-131072-windows-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 131,072 context \u00b7 C2 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 1799364544 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c2-131072-windows-calculated-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C2 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 1799364544 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c3-128k-int8", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 3 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 391118848, "full_context_requests": 3, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c3-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C3 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 2699046848 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c3-131072-windows-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 131,072 context \u00b7 C3 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 2699046848 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c3-131072-windows-calculated-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C3 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 2699046848 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c4-128k-int8", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 4 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 362807296, "full_context_requests": 4, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c4-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C4 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3598729152 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c4-131072-windows-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 131,072 context \u00b7 C4 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3598729152 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c4-131072-windows-calculated-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C4 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3598729152 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c5-128k-int8", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 5 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 363855872, "full_context_requests": 5, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c5-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C5 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 4498411456 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c5-131072-windows-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 131,072 context \u00b7 C5 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 4498411456 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c5-131072-windows-calculated-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C5 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 4498411456 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c6-128k-int8", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 6 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 368050176, "full_context_requests": 6, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c6-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C6 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 5398093760 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c6-131072-windows-calculated-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C6 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 5398093760 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c7-128k-int8", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 7 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 411041792, "full_context_requests": 7, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c7-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C7 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 6297776064 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c7-131072-windows-calculated-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C7 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 6297776064 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c8-128k-int8", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 8 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 383778816, "full_context_requests": 8, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c8-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C8 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 7197458368 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c8-131072-windows-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 131,072 context \u00b7 C8 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 7197458368 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 19776776704, "artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c8-131072-windows-calculated-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "groupwise-int-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C8 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 7197458368 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 } ], "min_vram_mib_per_gpu": 32606, "name": "Muse-Glimmer-30B INT (Q4 DFlash2)", "qualified_sm": [ "12.0" ], "sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66", "tp": 1, "url": "https://huggingface.co/SectileLabs/muse_glimmer_30b_int/resolve/84fa2b636d8b382ad670387024b6b69cdd5a22e5/muse_glimmer_30b_int_df2.ginfer" } ] }, { "description": "GInfer packages of Muse-Glimmer-30B in NVFP4. They run only on GInfer and GChat.", "developer": "SectileLabs", "downloads": 0, "library_name": "ginfer", "model_name": "SectileLabs/muse_glimmer_30b_nvfp4", "name": "Muse-Glimmer-30B NVFP4", "releases": [ { "bytes": 22447498752, "capabilities": [ "tools", "vision" ], "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "launch_profiles": [ { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c1-128k-nvfp4", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 1 request \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 808452096, "full_context_requests": 1, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c1-128k-nvfp4-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C1 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 534862848 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c1-128k-nvfp4-windows", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 1 request \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 415236096, "full_context_requests": 1, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c1-128k-nvfp4-windows-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C1 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 534862848 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c2-128k-nvfp4", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 2 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 822083584, "full_context_requests": 2, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c2-128k-nvfp4-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C2 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 1069725696 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c2-128k-nvfp4-windows", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 2 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 409993216, "full_context_requests": 2, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c2-128k-nvfp4-windows-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C2 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 1069725696 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c3-128k-nvfp4", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 3 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 822083584, "full_context_requests": 3, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c3-128k-nvfp4-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C3 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 1604588544 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c3-128k-nvfp4-windows", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 3 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 407896064, "full_context_requests": 3, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c3-128k-nvfp4-windows-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C3 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 1604588544 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c4-128k-nvfp4", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 4 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 822083584, "full_context_requests": 4, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c4-128k-nvfp4-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C4 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 2139451392 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c4-128k-nvfp4-windows", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 4 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 373293056, "full_context_requests": 4, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c4-128k-nvfp4-windows-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C4 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 2139451392 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c5-128k-nvfp4", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 5 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 796917760, "full_context_requests": 5, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c5-128k-nvfp4-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C5 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 2674314240 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c5-128k-nvfp4-windows", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 5 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 370147328, "full_context_requests": 5, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c5-128k-nvfp4-windows-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C5 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 2674314240 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c6-128k-nvfp4", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 6 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 791674880, "full_context_requests": 6, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c6-128k-nvfp4-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C6 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3209177088 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c6-128k-nvfp4-windows", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 6 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 376438784, "full_context_requests": 6, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c6-128k-nvfp4-windows-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C6 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3209177088 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c7-128k-nvfp4", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 7 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 792723456, "full_context_requests": 7, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c7-128k-nvfp4-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C7 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3744039936 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c7-128k-nvfp4-windows", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 7 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 328204288, "full_context_requests": 7, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c7-128k-nvfp4-windows-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C7 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3744039936 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c8-128k-nvfp4", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 8 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 788529152, "full_context_requests": 8, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c8-128k-nvfp4-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C8 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 4278902784 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c8-128k-nvfp4-windows", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 8 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 350224384, "full_context_requests": 8, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22447498752, "artifact_sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-nvfp4-tp1-c8-128k-nvfp4-windows-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4-dflash-nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C8 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 4278902784 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 } ], "min_vram_mib_per_gpu": 32606, "name": "Muse-Glimmer-30B NVFP4 (NVFP4 DFlash2)", "qualified_sm": [ "12.0" ], "sha256": "6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f", "tp": 1, "url": "https://huggingface.co/SectileLabs/muse_glimmer_30b_nvfp4/resolve/1b5641505de3936ba0bcd264788176dab47a1e7f/muse_glimmer_30b_nvfp4_df2.ginfer" }, { "bytes": 22360828416, "capabilities": [ "tools" ], "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4" }, "launch_profiles": [ { "artifact_bytes": 22360828416, "artifact_sha256": "878d6de4e0742f07fd7e10ca1efd819111a894325435d0297bf352ef72ae8d5c", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-tp1-c1-128k-int8", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 1 request", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 1783627776, "full_context_requests": 1, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22360828416, "artifact_sha256": "878d6de4e0742f07fd7e10ca1efd819111a894325435d0297bf352ef72ae8d5c", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-tp1-c1-128k-int8-windows", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 1 request", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 1932525568, "full_context_requests": 1, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22360828416, "artifact_sha256": "878d6de4e0742f07fd7e10ca1efd819111a894325435d0297bf352ef72ae8d5c", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-tp1-c2-128k-int8", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 2 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 1356857344, "full_context_requests": 2, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22360828416, "artifact_sha256": "878d6de4e0742f07fd7e10ca1efd819111a894325435d0297bf352ef72ae8d5c", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-tp1-c2-128k-int8-windows", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 2 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 1710227456, "full_context_requests": 2, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22360828416, "artifact_sha256": "878d6de4e0742f07fd7e10ca1efd819111a894325435d0297bf352ef72ae8d5c", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-tp1-c4-128k-int8", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 4 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 1426063360, "full_context_requests": 4, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22360828416, "artifact_sha256": "878d6de4e0742f07fd7e10ca1efd819111a894325435d0297bf352ef72ae8d5c", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-tp1-c4-128k-int8-windows", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 4 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 1907359744, "full_context_requests": 4, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 22360828416, "artifact_sha256": "878d6de4e0742f07fd7e10ca1efd819111a894325435d0297bf352ef72ae8d5c", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-nvfp4-dflash-tp1-c8-128k-int8", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "nvfp4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Muse Glimmer Text \u00b7 128K context \u00b7 8 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 1668284416, "full_context_requests": 8, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 } ], "min_vram_mib_per_gpu": 32607, "name": "Muse-Glimmer-30B NVFP4 (Q4 DFlash2)", "qualified_sm": [ "12.0" ], "sha256": "878d6de4e0742f07fd7e10ca1efd819111a894325435d0297bf352ef72ae8d5c", "tp": 1, "url": "https://huggingface.co/SectileLabs/muse_glimmer_30b_nvfp4/resolve/1b5641505de3936ba0bcd264788176dab47a1e7f/muse_glimmer_30b_nvfp4_df2q4.ginfer" } ] }, { "description": "GInfer packages of Muse-Glimmer-30B in SMOL 2-bit mix. They run only on GInfer and GChat. This package may repeat itself in long generations at higher context lengths. For long creative text, use the INT or NVFP4 package of Muse-Glimmer-30B where it fits your GPU.", "developer": "SectileLabs", "downloads": 0, "library_name": "ginfer", "model_name": "SectileLabs/muse_glimmer_30b_smol", "name": "Muse-Glimmer-30B SMOL 2-bit mix", "releases": [ { "bytes": 13838508544, "capabilities": [ "tools", "vision" ], "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "launch_profiles": [ { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "id": "rtx-pro6000-muse-smol-q2g64-mix-dflash-q4-tp1-c1-128k-int8-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 97887, "name": "Muse 30B 2-bit mix \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 1 request \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 87688216576, "required_kv_bytes_per_rank": 1042081280 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 315621376, "full_context_requests": 0, "smoke_requests": 1, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 96 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "id": "rtx-pro6000-muse-smol-q2g64-mix-dflash-q4-tp1-c1-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 97887, "name": "Muse 30B 2-bit mix Vision \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 1 request \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": true, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 86094381056, "required_kv_bytes_per_rank": 1042081280 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 315621376, "full_context_requests": 0, "smoke_requests": 1, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 96 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "id": "rtx-pro6000-muse-smol-q2g64-mix-dflash-q4-tp1-c2-128k-int8-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 97887, "name": "Muse 30B 2-bit mix \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 2 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 87455432704, "required_kv_bytes_per_rank": 2084162560 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 315621376, "full_context_requests": 0, "smoke_requests": 2, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 96 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "id": "rtx-pro6000-muse-smol-q2g64-mix-dflash-q4-tp1-c2-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 97887, "name": "Muse 30B 2-bit mix Vision \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 2 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": true, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 85769322496, "required_kv_bytes_per_rank": 2084162560 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 315621376, "full_context_requests": 0, "smoke_requests": 2, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 96 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "id": "rtx-pro6000-muse-smol-q2g64-mix-dflash-q4-tp1-c4-128k-int8-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 97887, "name": "Muse 30B 2-bit mix \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 4 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 87090528256, "required_kv_bytes_per_rank": 4168325120 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 315621376, "full_context_requests": 0, "smoke_requests": 4, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 96 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "id": "rtx-pro6000-muse-smol-q2g64-mix-dflash-q4-tp1-c4-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 97887, "name": "Muse 30B 2-bit mix Vision \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 4 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": true, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 85320531968, "required_kv_bytes_per_rank": 4168325120 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 315621376, "full_context_requests": 0, "smoke_requests": 4, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 96 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "id": "rtx-pro6000-muse-smol-q2g64-mix-dflash-q4-tp1-c8-128k-int8-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 97887, "name": "Muse 30B 2-bit mix \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 8 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 86585114624, "required_kv_bytes_per_rank": 8336650240 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 315621376, "full_context_requests": 0, "smoke_requests": 8, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 96 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "id": "rtx-pro6000-muse-smol-q2g64-mix-dflash-q4-tp1-c8-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 97887, "name": "Muse 30B 2-bit mix Vision \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 8 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": true, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 84448116736, "required_kv_bytes_per_rank": 8336650240 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 315621376, "full_context_requests": 0, "smoke_requests": 8, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 96 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "8.6", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 3080 Ti Laptop GPU", "id": "rtx3080ti-laptop-muse-smol-q2g64-mix-dflash-q4-tp1-c1-128k-int8-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 16384, "name": "Muse 30B 2-bit mix \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 1 request \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 2965372928, "required_kv_bytes_per_rank": 1042081280 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 316669952, "full_context_requests": 0, "smoke_requests": 1, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 16 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "8.6", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 3080 Ti Laptop GPU", "id": "rtx3080ti-laptop-muse-smol-q2g64-mix-dflash-q4-tp1-c1-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 16384, "name": "Muse 30B 2-bit mix Vision \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 1 request \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": true, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 1484783616, "required_kv_bytes_per_rank": 1042081280 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 316669952, "full_context_requests": 0, "smoke_requests": 1, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 16 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "8.6", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 3080 Ti Laptop GPU", "id": "rtx3080ti-laptop-muse-smol-q2g64-mix-dflash-q4-tp1-c2-128k-int8-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 16384, "name": "Muse 30B 2-bit mix \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 2 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 2715811840, "required_kv_bytes_per_rank": 2084162560 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 316669952, "full_context_requests": 0, "smoke_requests": 2, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 16 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "8.6", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 3080 Ti Laptop GPU", "id": "rtx3080ti-laptop-muse-smol-q2g64-mix-dflash-q4-tp1-c2-61k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 62464, "min_memory_mib_per_gpu": 16384, "name": "Muse 30B 2-bit mix Vision \u00b7 9d4ca1b2 \u00b7 61K context \u00b7 2 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": true, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 1142947840, "required_kv_bytes_per_rank": 1141442560 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 316669952, "full_context_requests": 0, "smoke_requests": 2, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 16 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "8.6", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 3080 Ti Laptop GPU", "id": "rtx3080ti-laptop-muse-smol-q2g64-mix-dflash-q4-tp1-c4-60k-int8-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 61440, "min_memory_mib_per_gpu": 16384, "name": "Muse 30B 2-bit mix \u00b7 9d4ca1b2 \u00b7 60K context \u00b7 4 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 2260729856, "required_kv_bytes_per_rank": 2254770176 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 316669952, "full_context_requests": 0, "smoke_requests": 4, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 16 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "8.6", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 3080 Ti Laptop GPU", "id": "rtx3080ti-laptop-muse-smol-q2g64-mix-dflash-q4-tp1-c7-16k-int8-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 16384, "min_memory_mib_per_gpu": 16384, "name": "Muse 30B 2-bit mix \u00b7 9d4ca1b2 \u00b7 16K context \u00b7 7 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 1797259264, "required_kv_bytes_per_rank": 1779305472 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 316669952, "full_context_requests": 0, "smoke_requests": 7, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 16 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "8.6", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 3080 Ti Laptop GPU", "id": "rtx3080ti-laptop-muse-smol-q2g64-mix-dflash-q4-tp1-c8-8k-int8-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 8192, "min_memory_mib_per_gpu": 16384, "name": "Muse 30B 2-bit mix \u00b7 9d4ca1b2 \u00b7 8K context \u00b7 8 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 1593835520, "required_kv_bytes_per_rank": 1582880768 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 316669952, "full_context_requests": 0, "smoke_requests": 8, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 16 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-glimmer-smol-q2g64-mix-dflash-q4-tp1-c1-128k-int8-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse 30B 2-bit mix \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 1 request \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 17379098624, "required_kv_bytes_per_rank": 1042081280 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 658505728, "full_context_requests": 0, "smoke_requests": 1, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-glimmer-smol-q2g64-mix-dflash-q4-tp1-c1-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse 30B 2-bit mix Vision \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 1 request \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": true, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 15814623232, "required_kv_bytes_per_rank": 1042081280 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 584056832, "full_context_requests": 0, "smoke_requests": 1, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-glimmer-smol-q2g64-mix-dflash-q4-tp1-c2-128k-int8-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse 30B 2-bit mix \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 2 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 17148411904, "required_kv_bytes_per_rank": 2084162560 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 652214272, "full_context_requests": 0, "smoke_requests": 2, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-glimmer-smol-q2g64-mix-dflash-q4-tp1-c2-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse 30B 2-bit mix Vision \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 2 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": true, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 15510536192, "required_kv_bytes_per_rank": 2084162560 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 632291328, "full_context_requests": 0, "smoke_requests": 2, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-glimmer-smol-q2g64-mix-dflash-q4-tp1-c4-128k-int8-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse 30B 2-bit mix \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 4 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 16844324864, "required_kv_bytes_per_rank": 4168325120 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 454033408, "full_context_requests": 0, "smoke_requests": 4, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-glimmer-smol-q2g64-mix-dflash-q4-tp1-c4-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse 30B 2-bit mix Vision \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 4 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": true, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 15013511168, "required_kv_bytes_per_rank": 4168325120 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 492830720, "full_context_requests": 0, "smoke_requests": 4, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-glimmer-smol-q2g64-mix-dflash-q4-tp1-c8-128k-int8-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse 30B 2-bit mix 9d4ca1b2 \u00b7 128K context \u00b7 8 requests \u00b7 INT8 KV \u00b7 calculated startup + smoke", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 15189450752, "required_kv_bytes_per_rank": 8336650240 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 328204288, "full_context_requests": 0, "smoke_requests": 8, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-glimmer-smol-q2g64-mix-dflash-q4-tp1-c8-128k-int8-vision-calculated", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse 30B 2-bit mix Vision \u00b7 9d4ca1b2 \u00b7 128K context \u00b7 8 requests \u00b7 calculated", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": true, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "calculation": { "available_kv_bytes_per_rank": 14145290240, "required_kv_bytes_per_rank": 8336650240 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 516947968, "full_context_requests": 0, "smoke_requests": 8, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 13838508544, "artifact_sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-muse-glimmer-smol-q2g64-mix-dflash-q4-tp1-c8-128k-int8-windows-screened", "identity": { "model_id": "muse-glimmer-30b", "weights_id": "smol-q2g64-mix-dflash-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Muse 30B 2-bit mix 9d4ca1b2 \u00b7 128K context \u00b7 8 requests \u00b7 INT8 KV \u00b7 screened", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, screened-startup tier", "free_bytes_per_gpu": 449839104, "full_context_requests": 0, "smoke_requests": 8, "tier": "screened-startup" }, "tp": 1, "vram_tier_gib": 32 } ], "min_vram_mib_per_gpu": 16384, "name": "Muse-Glimmer-30B SMOL 2-bit mix (Q4 DFlash2)", "qualified_sm": [ "12.0", "8.6" ], "sha256": "9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c", "tp": 1, "url": "https://huggingface.co/SectileLabs/muse_glimmer_30b_smol/resolve/3328ffe3067008c5d0ece5ccd72acf26f7bc407b/muse_glimmer_30b_smol_df2.ginfer" } ] }, { "description": "GInfer packages of Qwen3.6-35B-A3B in INT. They run only on GInfer and GChat.", "developer": "SectileLabs", "downloads": 0, "library_name": "ginfer", "model_name": "SectileLabs/qwen36_35b_a3b_int", "name": "Qwen3.6-35B-A3B INT", "releases": [ { "bytes": 20016414464, "capabilities": [ "tools" ], "identity": { "model_id": "qwen3.6-35b-a3b", "weights_id": "groupwise-int" }, "launch_profiles": [ { "artifact_bytes": 20016414464, "artifact_sha256": "daf27aafd5f461f81de9a7fa63b8c3f1e56c3ac42b2da858842275879b90bdfa", "concurrency": 8, "cpu": { "isa": "amx", "min_ram_mib": 92160 }, "draft_tp": 0, "id": "cpu-amx-qwen36-35b-a3b-groupwise-int-tp1-c8-256k-int8-screened", "identity": { "model_id": "qwen3.6-35b-a3b", "weights_id": "groupwise-int" }, "max_context": 262144, "name": "Qwen3.6 35B-A3B INT on the CPU (AMX, 96 GB RAM class): 8 replies at once, 262,144 tokens; screened, not full-context qualified", "options": { "cpu_ram_headroom_gib": null, "device": "cpu", "draft_policy": "auto", "draft_tokens": 0, "draft_tp": 0, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_dtype": "int8", "spec": "none", "structured_outputs": true, "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, screened-startup tier", "full_context_requests": 0, "host_free_bytes": 705639821312, "smoke_requests": 8, "tier": "screened-startup" }, "tp": 1 }, { "artifact_bytes": 20016414464, "artifact_sha256": "daf27aafd5f461f81de9a7fa63b8c3f1e56c3ac42b2da858842275879b90bdfa", "concurrency": 4, "cpu": { "isa": "avx512", "min_ram_mib": 61440 }, "draft_tp": 0, "id": "cpu-avx512-qwen36-35b-a3b-groupwise-int-tp1-c4-256k-int8-screened", "identity": { "model_id": "qwen3.6-35b-a3b", "weights_id": "groupwise-int" }, "max_context": 262144, "name": "Qwen3.6 35B-A3B INT on the CPU (AVX-512, 64 GB RAM class): 4 replies at once, 262,144 tokens; screened, not full-context qualified", "options": { "cpu_ram_headroom_gib": null, "device": "cpu", "draft_policy": "auto", "draft_tokens": 0, "draft_tp": 0, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_dtype": "int8", "spec": "none", "structured_outputs": true, "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, screened-startup tier", "full_context_requests": 0, "host_free_bytes": 58593513472, "smoke_requests": 4, "tier": "screened-startup" }, "tp": 1 }, { "artifact_bytes": 20016414464, "artifact_sha256": "daf27aafd5f461f81de9a7fa63b8c3f1e56c3ac42b2da858842275879b90bdfa", "concurrency": 4, "cpu": { "isa": "avx512", "min_ram_mib": 61440 }, "draft_tp": 0, "id": "cpu-avx512-qwen36-35b-a3b-groupwise-int-tp1-c4-256k-int8-windows-screened", "identity": { "model_id": "qwen3.6-35b-a3b", "weights_id": "groupwise-int" }, "max_context": 262144, "name": "Qwen3.6 35B-A3B INT on the CPU, Windows (AVX-512, 64 GB RAM class): 4 replies at once, 262,144 tokens; screened, not full-context qualified", "options": { "cpu_ram_headroom_gib": null, "device": "cpu", "draft_policy": "auto", "draft_tokens": 0, "draft_tp": 0, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_dtype": "int8", "spec": "none", "structured_outputs": true, "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, screened-startup tier", "full_context_requests": 0, "host_free_bytes": 30502268928, "smoke_requests": 4, "tier": "screened-startup" }, "tp": 1 } ], "min_vram_mib_per_gpu": 0, "name": "Qwen3.6-35B-A3B INT", "qualified_cpu": [ "amx", "avx512" ], "qualified_sm": [], "sha256": "daf27aafd5f461f81de9a7fa63b8c3f1e56c3ac42b2da858842275879b90bdfa", "tp": 1, "url": "https://huggingface.co/SectileLabs/qwen36_35b_a3b_int/resolve/1f79f4bb61cc864ee0fbe0801870db9654829972/qwen36_35b_a3b_int_rtn.ginfer" } ] }, { "description": "GInfer packages of Qwen3.6-35B-A3B in SMOL 2-bit mix. They run only on GInfer and GChat.", "developer": "SectileLabs", "downloads": 0, "library_name": "ginfer", "model_name": "SectileLabs/qwen36_35b_a3b_smol", "name": "Qwen3.6-35B-A3B SMOL 2-bit mix", "releases": [ { "bytes": 14446362368, "capabilities": [ "tools" ], "identity": { "model_id": "qwen3.6-35b-a3b", "weights_id": "smol-q2g64" }, "launch_profiles": [ { "artifact_bytes": 14446362368, "artifact_sha256": "a7af668a8b8f63f37a0951efd944f38777fb30992c65b09abd9da321557be8a9", "concurrency": 1, "cpu": { "isa": "amx", "min_ram_mib": 22528 }, "draft_tp": 0, "id": "cpu-amx-qwen36-35b-a3b-smol-q2g64-tp1-c1-64k-int8-screened", "identity": { "model_id": "qwen3.6-35b-a3b", "weights_id": "smol-q2g64" }, "max_context": 65536, "name": "Qwen3.6 35B-A3B SMOL on the CPU (AMX, 24 GB RAM class): 1 reply at once, 65,536 tokens; screened, not full-context qualified", "options": { "cpu_ram_headroom_gib": null, "device": "cpu", "draft_policy": "auto", "draft_tokens": 0, "draft_tp": 0, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_dtype": "int8", "spec": "none", "structured_outputs": true, "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, screened-startup tier", "full_context_requests": 0, "host_free_bytes": 776355155968, "smoke_requests": 1, "tier": "screened-startup" }, "tp": 1 }, { "artifact_bytes": 14446362368, "artifact_sha256": "a7af668a8b8f63f37a0951efd944f38777fb30992c65b09abd9da321557be8a9", "concurrency": 1, "cpu": { "isa": "avx512", "min_ram_mib": 22528 }, "draft_tp": 0, "id": "cpu-avx512-qwen36-35b-a3b-smol-q2g64-tp1-c1-64k-int8-screened", "identity": { "model_id": "qwen3.6-35b-a3b", "weights_id": "smol-q2g64" }, "max_context": 65536, "name": "Qwen3.6 35B-A3B SMOL on the CPU (AVX-512, 24 GB RAM class): 1 reply at once, 65,536 tokens; screened, not full-context qualified", "options": { "cpu_ram_headroom_gib": null, "device": "cpu", "draft_policy": "auto", "draft_tokens": 0, "draft_tp": 0, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_dtype": "int8", "spec": "none", "structured_outputs": true, "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, screened-startup tier", "full_context_requests": 0, "host_free_bytes": 85359411200, "smoke_requests": 1, "tier": "screened-startup" }, "tp": 1 }, { "artifact_bytes": 14446362368, "artifact_sha256": "a7af668a8b8f63f37a0951efd944f38777fb30992c65b09abd9da321557be8a9", "concurrency": 1, "cpu": { "isa": "avx512", "min_ram_mib": 22528 }, "draft_tp": 0, "id": "cpu-avx512-qwen36-35b-a3b-smol-q2g64-tp1-c1-64k-int8-windows-screened", "identity": { "model_id": "qwen3.6-35b-a3b", "weights_id": "smol-q2g64" }, "max_context": 65536, "name": "Qwen3.6 35B-A3B SMOL on the CPU, Windows (AVX-512, 24 GB RAM class): 1 reply at once, 65,536 tokens; screened, not full-context qualified", "options": { "cpu_ram_headroom_gib": null, "device": "cpu", "draft_policy": "auto", "draft_tokens": 0, "draft_tp": 0, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_dtype": "int8", "spec": "none", "structured_outputs": true, "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, screened-startup tier", "full_context_requests": 0, "host_free_bytes": 57471258624, "smoke_requests": 1, "tier": "screened-startup" }, "tp": 1 } ], "min_vram_mib_per_gpu": 0, "name": "Qwen3.6-35B-A3B SMOL 2-bit mix", "qualified_cpu": [ "amx", "avx512" ], "qualified_sm": [], "sha256": "a7af668a8b8f63f37a0951efd944f38777fb30992c65b09abd9da321557be8a9", "tp": 1, "url": "https://huggingface.co/SectileLabs/qwen36_35b_a3b_smol/resolve/93a0b14cbcf080e73a4b3f86c86623976721d1b3/qwen36_35b_a3b_smol_autoround_down37.ginfer" } ] }, { "description": "GInfer packages of Qwen3.8-27B in INT. They run only on GInfer and GChat.", "developer": "SectileLabs", "downloads": 0, "library_name": "ginfer", "model_name": "SectileLabs/qwen38_27b_int", "name": "Qwen3.8-27B INT", "releases": [ { "bytes": 18424985856, "capabilities": [ "tools", "vision" ], "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "launch_profiles": [ { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c1-256k-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 262144, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 256K context \u00b7 1 request \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 359661568, "full_context_requests": 1, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c1-256k-int8-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 262144, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 262,144 context \u00b7 C1 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 8858370048 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c1-262144-windows-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 262144, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Text \u00b7 262,144 context \u00b7 C1 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 8858370048 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c1-262144-windows-calculated-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 262144, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 262,144 context \u00b7 C1 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 8858370048 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c2-176128-windows-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 176128, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Text \u00b7 176,128 context \u00b7 C2 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 11903434752 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c2-176128-windows-calculated-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 131650, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 131,650 context \u00b7 C2 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 8897433600 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c2-184071-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 184071, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 184,071-token context \u00b7 2 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 335544320, "full_context_requests": 2, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c2-184071-int8-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 139594, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 139,594 context \u00b7 C2 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 9434320896 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c3-110964-windows-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 110964, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Text \u00b7 110,964 context \u00b7 C3 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 11249086464 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c3-110964-windows-calculated-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 78209, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 78,209 context \u00b7 C3 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 7928515584 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c3-116260-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 116260, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 116,260-token context \u00b7 3 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 330301440, "full_context_requests": 3, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c3-116260-int8-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 83505, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 83,505 context \u00b7 C3 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 8465402880 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c4-79701-windows-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 79701, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Text \u00b7 79,701 context \u00b7 C4 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 10773024768 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c4-79701-windows-calculated-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 52808, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 52,808 context \u00b7 C4 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 7137951744 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c4-83673-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 83673, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 83,673-token context \u00b7 4 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 330301440, "full_context_requests": 4, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c4-83673-int8-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 56779, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 56,779 context \u00b7 C4 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 7674703872 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c5-61266-windows-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 61266, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Text \u00b7 61,266 context \u00b7 C5 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 10351503360 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c5-61266-windows-calculated-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 37888, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 37,888 context \u00b7 C5 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 6401556480 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c5-64443-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 64443, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 64,443-token context \u00b7 5 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 397410304, "full_context_requests": 5, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c5-64443-int8-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 41067, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 41,067 context \u00b7 C5 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 6938680320 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c6-49772-windows-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 49772, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Text \u00b7 49,772 context \u00b7 C6 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 10091372544 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c6-49772-windows-calculated-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 29511, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 29,511 context \u00b7 C6 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 5983414272 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c6-52420-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 52420, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 52,420-token context \u00b7 6 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 319815680, "full_context_requests": 6, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c6-52420-int8-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 31388, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 31,388 context \u00b7 C6 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 6363979776 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c7-41695-windows-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 41695, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Text \u00b7 41,695 context \u00b7 C7 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 9862702080 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c7-41695-windows-calculated-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 25446, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 25,446 context \u00b7 C7 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 6019098624 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c7-43965-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 43965, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 43,965-token context \u00b7 7 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 395313152, "full_context_requests": 7, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c7-43965-int8-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 26874, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 26,874 context \u00b7 C7 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 6356883456 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c8-35459-windows-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 35459, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Text \u00b7 35,459 context \u00b7 C8 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 9585844224 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c8-35459-windows-calculated-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 22119, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 22,119 context \u00b7 C8 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 5979561984 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c8-37445-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 37445, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 37,445-token context \u00b7 8 requests \u00b7 INT8 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 328204288, "full_context_requests": 8, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 18424985856, "artifact_sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-groupwise-int-dflash2-q4-tp1-c8-37445-int8-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "groupwise-int-dflash2-q4" }, "max_context": 23398, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 23,398 context \u00b7 C8 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 6325321728 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 } ], "min_vram_mib_per_gpu": 32606, "name": "Qwen3.8-27B INT (Q4 DFlash2)", "qualified_sm": [ "12.0" ], "sha256": "bcf29abd1598f5666c641da7c59ac28bd999aa98a8d6ba533ae7aa579abd06a7", "tp": 1, "url": "https://huggingface.co/SectileLabs/qwen38_27b_int/resolve/705cf778a8d9d4ef4defdb7ebfecb4356d76f661/qwen38_27b_int_df2.ginfer" } ] }, { "description": "GInfer packages of Qwen3.8-27B in NVFP4. They run only on GInfer and GChat.", "developer": "SectileLabs", "downloads": 0, "library_name": "ginfer", "model_name": "SectileLabs/qwen38_27b_nvfp4", "name": "Qwen3.8-27B NVFP4", "releases": [ { "bytes": 21767307520, "capabilities": [ "tools", "vision" ], "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "launch_profiles": [ { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c1-256k-nvfp4", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 262144, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 256K context \u00b7 1 request \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 601882624, "full_context_requests": 1, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c1-256k-nvfp4-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 262144, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 262,144 context \u00b7 C1 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 4957667328 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c1-262144-windows", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 262144, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 Text \u00b7 262,144-token context \u00b7 1 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "available_kv_bytes_per_rank": 10077863936, "required_kv_bytes_per_rank": 4957667328 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 418381824, "full_context_requests": 0, "smoke_requests": 1, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c1-262144-windows-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 262144, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 27B Vision + DFlash \u00b7 262,144 context \u00b7 C1 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 4957667328 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c2-224k-nvfp4", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 229376, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 224K context \u00b7 2 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 596639744, "full_context_requests": 2, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c2-224k-nvfp4-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 150528, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 150,528 context \u00b7 C2 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 5693571072 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c2-237568-windows", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 237568, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 Text \u00b7 237,568-token context \u00b7 2 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "available_kv_bytes_per_rank": 9650044928, "required_kv_bytes_per_rank": 8985772032 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 433061888, "full_context_requests": 0, "smoke_requests": 2, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c2-237568-windows-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 157696, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 27B Vision + DFlash \u00b7 157,696 context \u00b7 C2 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 5964693504 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c3-145k-nvfp4", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 148480, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 145K context \u00b7 3 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 612368384, "full_context_requests": 3, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c3-145k-nvfp4-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 89088, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 89,088 context \u00b7 C3 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 5054496768 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c3-152576-windows", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 152576, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 Text \u00b7 152,576-token context \u00b7 3 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "available_kv_bytes_per_rank": 9366929408, "required_kv_bytes_per_rank": 8656551936 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-startup-smoke tier", "free_bytes_per_gpu": 319815680, "full_context_requests": 0, "smoke_requests": 3, "tier": "calculated-startup-smoke" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 3, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c3-152576-windows-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 94208, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 27B Vision + DFlash \u00b7 94,208 context \u00b7 C3 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 5344985088 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c4-104k-nvfp4", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 106496, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 104K context \u00b7 4 requests \u00b7 NVFP4 KV", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 612368384, "full_context_requests": 4, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c4-104k-nvfp4-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 58368, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 58,368 context \u00b7 C4 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 4415422464 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c4-99328-windows-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 99328, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Text \u00b7 99,328 context \u00b7 C4 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 7513964544 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c4-99328-windows-calculated-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 51200, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 51,200 context \u00b7 C4 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3873177600 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c5-75776-windows-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 75776, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Text \u00b7 75,776 context \u00b7 C5 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 7165378560 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c5-75776-windows-calculated-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 33792, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 33,792 context \u00b7 C5 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3195371520 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c5-79k-nvfp4", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 80896, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 79K context \u00b7 5 requests \u00b7 NVFP4 KV \u00b7 mixed graph/eager", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 612368384, "full_context_requests": 5, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 5, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c5-79k-nvfp4-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 39936, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 39,936 context \u00b7 C5 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3776348160 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c6-60416-windows-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 60416, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Text \u00b7 60,416 context \u00b7 C6 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 6855524352 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c6-60416-windows-calculated-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 26624, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 26,624 context \u00b7 C6 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3021078528 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c6-63k-nvfp4", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 64512, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 63K context \u00b7 6 requests \u00b7 NVFP4 KV \u00b7 mixed graph/eager", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 605028352, "full_context_requests": 6, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 6, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c6-63k-nvfp4-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 28948, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 28,948 context \u00b7 C6 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3369664512 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c7-48128-windows-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 48128, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Text \u00b7 48,128 context \u00b7 C7 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 6371377152 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c7-48128-windows-calculated-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 22528, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 22,528 context \u00b7 C7 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 2982346752 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c7-51k-nvfp4", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 52224, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 51K context \u00b7 7 requests \u00b7 NVFP4 KV \u00b7 mixed graph/eager", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 715128832, "full_context_requests": 7, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 7, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c7-51k-nvfp4-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 24572, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 24,572 context \u00b7 C7 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3253469184 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c8-39936-windows-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 39936, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Text \u00b7 39,936 context \u00b7 C8 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 6042157056 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c8-39936-windows-calculated-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 18781, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 18,781 context \u00b7 C8 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "windows", "qualification": { "calculation": { "required_kv_bytes_per_rank": 2943614976 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c8-42k-nvfp4", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 43008, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 Text \u00b7 42K context \u00b7 8 requests \u00b7 NVFP4 KV \u00b7 mixed graph/eager", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 715128832, "full_context_requests": 8, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21767307520, "artifact_sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-nvfp4-tp1-c8-42k-nvfp4-vision-calculated", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-nvfp4" }, "max_context": 20480, "min_memory_mib_per_gpu": 32606, "name": "Qwen3.8 27B Vision + DFlash \u00b7 20,480 context \u00b7 C8 \u00b7 calculated", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 524288000, "kv_dtype": "nvfp4", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": true }, "platform": "linux", "qualification": { "calculation": { "required_kv_bytes_per_rank": 3098542080 }, "engine_revision": "v1", "evidence": "GInfer v1 release qualification, calculated-pending-validation tier", "full_context_requests": 0, "tier": "calculated-pending-validation" }, "tp": 1, "vram_tier_gib": 32 } ], "min_vram_mib_per_gpu": 32606, "name": "Qwen3.8-27B NVFP4 (NVFP4 DFlash2)", "qualified_sm": [ "12.0" ], "sha256": "ad4972ce2903d4b16ba6d966ee86a77b8816778832a9d176c26be139de2b77e0", "tp": 1, "url": "https://huggingface.co/SectileLabs/qwen38_27b_nvfp4/resolve/9daa5650ad7518c0d0d0cd98a7c28661c3c4b616/qwen38_27b_nvfp4_df2.ginfer" }, { "bytes": 21707145472, "capabilities": [ "tools" ], "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-q4" }, "launch_profiles": [ { "artifact_bytes": 21707145472, "artifact_sha256": "54638e003c17d78f5c049f76d58b24c649786dc9c8b6165ba25b7ab7fe00dd21", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-q4-tp1-c1-128k-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 Text \u00b7 128K context \u00b7 1 request", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 2062548992, "full_context_requests": 1, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21707145472, "artifact_sha256": "54638e003c17d78f5c049f76d58b24c649786dc9c8b6165ba25b7ab7fe00dd21", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-q4-tp1-c2-64k-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-q4" }, "max_context": 65536, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 Text \u00b7 64K context \u00b7 2 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 1731198976, "full_context_requests": 2, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21707145472, "artifact_sha256": "54638e003c17d78f5c049f76d58b24c649786dc9c8b6165ba25b7ab7fe00dd21", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-q4-tp1-c4-32k-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-q4" }, "max_context": 32768, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 Text \u00b7 32K context \u00b7 4 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 1803550720, "full_context_requests": 4, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21707145472, "artifact_sha256": "54638e003c17d78f5c049f76d58b24c649786dc9c8b6165ba25b7ab7fe00dd21", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-nvfp4-dflash2-q4-tp1-c8-16k-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-q4" }, "max_context": 16384, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 Text \u00b7 16K context \u00b7 8 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 1757413376, "full_context_requests": 8, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21707145472, "artifact_sha256": "54638e003c17d78f5c049f76d58b24c649786dc9c8b6165ba25b7ab7fe00dd21", "compute_capability": "12.0", "concurrency": 1, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "windows-rtx5090-qwen38-nvfp4-dflash2-q4-tp1-c1-128k-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 Text \u00b7 128K context \u00b7 1 request", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 2198863872, "full_context_requests": 1, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21707145472, "artifact_sha256": "54638e003c17d78f5c049f76d58b24c649786dc9c8b6165ba25b7ab7fe00dd21", "compute_capability": "12.0", "concurrency": 2, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "windows-rtx5090-qwen38-nvfp4-dflash2-q4-tp1-c2-64k-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-q4" }, "max_context": 65536, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 Text \u00b7 64K context \u00b7 2 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 1774190592, "full_context_requests": 2, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21707145472, "artifact_sha256": "54638e003c17d78f5c049f76d58b24c649786dc9c8b6165ba25b7ab7fe00dd21", "compute_capability": "12.0", "concurrency": 4, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "windows-rtx5090-qwen38-nvfp4-dflash2-q4-tp1-c4-32k-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-q4" }, "max_context": 32768, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 Text \u00b7 32K context \u00b7 4 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 1816133632, "full_context_requests": 4, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 21707145472, "artifact_sha256": "54638e003c17d78f5c049f76d58b24c649786dc9c8b6165ba25b7ab7fe00dd21", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "windows-rtx5090-qwen38-nvfp4-dflash2-q4-tp1-c8-16k-int8", "identity": { "model_id": "qwen3.8-27b", "weights_id": "nvfp4-dflash2-q4" }, "max_context": 16384, "min_memory_mib_per_gpu": 32607, "name": "Qwen3.8 Text \u00b7 16K context \u00b7 8 requests", "options": { "draft_tokens": 4, "draft_tp": 1, "host_kv_cache_bytes": 0, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "spec": "dflash", "vision": false }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, full-context-tested tier", "free_bytes_per_gpu": 1845493760, "full_context_requests": 8, "tier": "full-context-tested" }, "tp": 1, "vram_tier_gib": 32 } ], "min_vram_mib_per_gpu": 32607, "name": "Qwen3.8-27B NVFP4 (Q4 DFlash2)", "qualified_sm": [ "12.0" ], "sha256": "54638e003c17d78f5c049f76d58b24c649786dc9c8b6165ba25b7ab7fe00dd21", "tp": 1, "url": "https://huggingface.co/SectileLabs/qwen38_27b_nvfp4/resolve/9daa5650ad7518c0d0d0cd98a7c28661c3c4b616/qwen38_27b_nvfp4_df2q4.ginfer" } ] }, { "description": "GInfer packages of Qwen3.8-27B in SMOL 2-bit mix. They run only on GInfer and GChat.", "developer": "SectileLabs", "downloads": 0, "library_name": "ginfer", "model_name": "SectileLabs/qwen38_27b_smol", "name": "Qwen3.8-27B SMOL 2-bit mix", "releases": [ { "bytes": 10667321600, "capabilities": [ "tools" ], "identity": { "model_id": "qwen3.8-27b", "weights_id": "smol-q2g64-mix-dflash2-q4" }, "launch_profiles": [ { "artifact_bytes": 10667321600, "artifact_sha256": "483343c28912ae5db3d00478f16fffb10dcbb0c1317f8c90c09c396fcb6e4ca4", "compute_capability": "8.6", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 3080 Ti Laptop GPU", "id": "rtx3080ti-laptop-qwen38-smol-q2g64-mix-dflash2-q4-tp1-c8-74k-int8-screened", "identity": { "model_id": "qwen3.8-27b", "weights_id": "smol-q2g64-mix-dflash2-q4" }, "max_context": 75776, "min_memory_mib_per_gpu": 16384, "name": "Qwen 27B 2-bit mix \u00b7 483343c2 \u00b7 74K context \u00b7 8 requests \u00b7 screened", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 1024, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, screened-startup tier", "free_bytes_per_gpu": 316555264, "full_context_requests": 0, "smoke_requests": 8, "tier": "screened-startup" }, "tp": 1, "vram_tier_gib": 16 }, { "artifact_bytes": 10667321600, "artifact_sha256": "483343c28912ae5db3d00478f16fffb10dcbb0c1317f8c90c09c396fcb6e4ca4", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-smol-q2g64-mix-dflash2-q4-tp1-c8-128k-int8-screened", "identity": { "model_id": "qwen3.8-27b", "weights_id": "smol-q2g64-mix-dflash2-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Qwen 27B 2-bit mix 483343c2 \u00b7 128K context \u00b7 8 requests \u00b7 INT8 KV \u00b7 screened", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 0, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, screened-startup tier", "free_bytes_per_gpu": 448790528, "full_context_requests": 0, "smoke_requests": 8, "tier": "screened-startup" }, "tp": 1, "vram_tier_gib": 32 }, { "artifact_bytes": 10667321600, "artifact_sha256": "483343c28912ae5db3d00478f16fffb10dcbb0c1317f8c90c09c396fcb6e4ca4", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 1, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-qwen38-smol-q2g64-mix-dflash2-q4-tp1-c8-128k-int8-windows-screened", "identity": { "model_id": "qwen3.8-27b", "weights_id": "smol-q2g64-mix-dflash2-q4" }, "max_context": 131072, "min_memory_mib_per_gpu": 32606, "name": "Qwen 27B 2-bit mix 483343c2 \u00b7 128K context \u00b7 8 requests \u00b7 INT8 KV \u00b7 screened", "options": { "draft_policy": "fixed", "draft_tokens": 7, "draft_tp": 1, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 0, "progressive_ready": false, "spec": "dflash", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "windows", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, screened-startup tier", "free_bytes_per_gpu": 373293056, "full_context_requests": 0, "smoke_requests": 8, "tier": "screened-startup" }, "tp": 1, "vram_tier_gib": 32 } ], "min_vram_mib_per_gpu": 16384, "name": "Qwen3.8-27B SMOL 2-bit mix (Q4 DFlash2)", "qualified_sm": [ "12.0", "8.6" ], "sha256": "483343c28912ae5db3d00478f16fffb10dcbb0c1317f8c90c09c396fcb6e4ca4", "tp": 1, "url": "https://huggingface.co/SectileLabs/qwen38_27b_smol/resolve/afb206f06630b2a7c5b43593fdc48b74d1a367a9/qwen38_27b_smol_df2.ginfer" } ] }, { "description": "GInfer packages of Qwen3.8-Flash-Next 180B in SMOL pure 2-bit. They run only on GInfer and GChat.", "developer": "SectileLabs", "downloads": 0, "library_name": "ginfer", "model_name": "SectileLabs/qwen38_flash_next_180b_smol", "name": "Qwen3.8-Flash-Next 180B SMOL pure 2-bit", "releases": [ { "bytes": 70247086080, "capabilities": [ "tools" ], "identity": { "model_id": "qwen3.8-flash-next", "weights_id": "smol-q2g64" }, "launch_profiles": [ { "artifact_bytes": 70247086080, "artifact_sha256": "8bbb85bf6ba669157045bac0a6fdd623dd87694b7b1fe09b7f2150fc3636ab08", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 0, "gpu_name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "id": "rtx-pro6000-flash-next-smol-q2g64-tp1-c8-128k-int8-screened", "identity": { "model_id": "qwen3.8-flash-next", "weights_id": "smol-q2g64" }, "max_context": 131072, "min_memory_mib_per_gpu": 97887, "name": "Flash Next pure 2-bit \u00b7 128K context \u00b7 8 requests \u00b7 INT8 KV \u00b7 screened", "options": { "draft_policy": "fixed", "draft_tokens": 3, "draft_tp": 0, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 1073741824, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 0, "progressive_ready": false, "spec": "mtp", "structured_outputs": true, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, screened-startup tier", "free_bytes_per_gpu": 18581880832, "full_context_requests": 0, "smoke_requests": 8, "tier": "screened-startup" }, "tp": 1, "vram_tier_gib": 96 }, { "artifact_bytes": 70247086080, "artifact_sha256": "8bbb85bf6ba669157045bac0a6fdd623dd87694b7b1fe09b7f2150fc3636ab08", "compute_capability": "8.6", "concurrency": 8, "draft_tp": 0, "gpu_name": "NVIDIA GeForce RTX 3080 Ti Laptop GPU", "id": "rtx3080ti-laptop-flash-next-smol-q2g64-tp1-c8-66k-int8-screened", "identity": { "model_id": "qwen3.8-flash-next", "weights_id": "smol-q2g64" }, "max_context": 67584, "min_memory_mib_per_gpu": 16384, "name": "Flash Next pure 2-bit 8bbb85bf \u00b7 66K context \u00b7 8 requests \u00b7 INT8 KV \u00b7 screened", "options": { "draft_policy": "fixed", "draft_tokens": 3, "draft_tp": 0, "expert_lfu": true, "flash_host_ram_headroom_gib": 10, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 0, "progressive_ready": false, "spec": "mtp", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, screened-startup tier", "free_bytes_per_gpu": 1459617792, "full_context_requests": 0, "smoke_requests": 8, "tier": "screened-startup" }, "tp": 1, "vram_tier_gib": 16 }, { "artifact_bytes": 70247086080, "artifact_sha256": "8bbb85bf6ba669157045bac0a6fdd623dd87694b7b1fe09b7f2150fc3636ab08", "compute_capability": "12.0", "concurrency": 8, "draft_tp": 0, "gpu_name": "NVIDIA GeForce RTX 5090", "id": "rtx5090-flash-next-smol-q2g64-tp1-c8-256k-int8-screened", "identity": { "model_id": "qwen3.8-flash-next", "weights_id": "smol-q2g64" }, "max_context": 262144, "min_memory_mib_per_gpu": 32606, "name": "Flash Next pure 2-bit 8bbb85bf \u00b7 256K context \u00b7 8 requests \u00b7 INT8 KV \u00b7 screened", "options": { "draft_policy": "fixed", "draft_tokens": 3, "draft_tp": 0, "expert_lfu": true, "flash_host_ram_headroom_gib": 10, "host_kv_cache_bytes": null, "kv_arena_bytes": null, "kv_arena_headroom_bytes": 314572800, "kv_dtype": "int8", "no_cuda_graph": false, "prefill_chunk": 0, "progressive_ready": false, "spec": "mtp", "structured_outputs": false, "vision": false, "vram_budget_bytes": null }, "platform": "linux", "qualification": { "engine_revision": "v1", "evidence": "GInfer v1 release qualification, screened-startup tier", "free_bytes_per_gpu": 1051721728, "full_context_requests": 0, "smoke_requests": 8, "tier": "screened-startup" }, "tp": 1, "vram_tier_gib": 32 } ], "min_vram_mib_per_gpu": 16384, "name": "Qwen3.8-Flash-Next 180B SMOL pure 2-bit", "qualified_sm": [ "12.0", "8.6" ], "sha256": "8bbb85bf6ba669157045bac0a6fdd623dd87694b7b1fe09b7f2150fc3636ab08", "tp": 1, "url": "https://huggingface.co/SectileLabs/qwen38_flash_next_180b_smol/resolve/10c9325d5630989b5baef0245c98d9bd3cc56c2d/qwen38_flash_next_180b_smol_mtp.ginfer" } ] } ], "schema_version": 1, "updated_at": "2026-10-30T17:38:00Z" }
{
"manifest_version": 1,
"models": [
{
"description": "GInfer packages of Muse-Glimmer-30B in INT. They run only on GInfer and GChat.",
"developer": "SectileLabs",
"downloads": 0,
"library_name": "ginfer",
"model_name": "SectileLabs/muse_glimmer_30b_int",
"name": "Muse-Glimmer-30B INT",
"releases": [
{
"bytes": 19776776704,
"capabilities": [
"tools",
"vision"
],
"identity": {
"model_id": "muse-glimmer-30b",
"weights_id": "groupwise-int-dflash-q4"
},
"launch_profiles": [
{
"artifact_bytes": 19776776704,
"artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66",
"compute_capability": "12.0",
"concurrency": 1,
"draft_tp": 1,
"gpu_name": "NVIDIA GeForce RTX 5090",
"id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c1-128k-int8",
"identity": {
"model_id": "muse-glimmer-30b",
"weights_id": "groupwise-int-dflash-q4"
},
"max_context": 131072,
"min_memory_mib_per_gpu": 32606,
"name": "Muse Glimmer Text \u00b7 128K context \u00b7 1 request \u00b7 INT8 KV",
"options": {
"draft_tokens": 4,
"draft_tp": 1,
"host_kv_cache_bytes": 0,
"kv_arena_bytes": null,
"kv_arena_headroom_bytes": 314572800,
"kv_dtype": "int8",
"no_cuda_graph": false,
"prefill_chunk": 1024,
"spec": "dflash",
"vision": false
},
"platform": "linux",
"qualification": {
"engine_revision": "v1",
"evidence": "GInfer v1 release qualification, full-context-tested tier",
"free_bytes_per_gpu": 458227712,
"full_context_requests": 1,
"tier": "full-context-tested"
},
"tp": 1,
"vram_tier_gib": 32
},
{
"artifact_bytes": 19776776704,
"artifact_sha256": "46b5d54ee64ccd278f2a1a4d2c6d5aa9fc9097e8db0c3d165179aee5f2cb5b66",
"compute_capability": "12.0",
"concurrency": 1,
"draft_tp": 1,
"gpu_name": "NVIDIA GeForce RTX 5090",
"id": "rtx5090-muse-groupwise-int-dflash-q4-tp1-c1-128k-int8-vision-calculated",
"identity": {
"model_id": "muse-glimmer-30b",
"weights_id": "groupwise-int-dflash-q4"
},
"max_context": 131072,
"min_memory_mib_per_gpu": 32606,
"name": "Muse Glimmer Vision + DFlash \u00b7 131,072 context \u00b7 C1 \u00b7 calculated",
"options": {
"draft_tokens": 4,
"draft_tp": 1,
"host_kv_cache_bytes": 0,
"kv_arena_bytes": null,
"kv_arena_headroom_bytes": 314572800,
"kv_dtype": "int8",
"no_cuda_graph": false,
"prefill_chunk": 1024,
"spec": "dflash",
"vision": true
},
"platform": "linux",
"qualification": {
"calculation": {
"required_kv_bytes_per_rank": 899682240
},
"engine_revision": "v1",
"evidence": "GInfer v1 release qualification, calculated-pending-validation tier",
"full_context_requests": 0,
"tier": "calculated-pending-validation"
},
"tp": 1,
"vram_tier_gib": 32
},