Temas / Models & quants
Models & quants
Auto topic: Models & quants
Issues
- #6442 [Engine] Inkling-2-Mini support (looped MoE from Thinking Machines)
- #6541 [Engine] strata-hy4 0.1.64: EXCEPTION_ILLEGAL_INSTRUCTION (AVX-512 EVEX) on Arrow Lake-HX — current workaround disables the 1.5x AVX-VNNI cold-expert kernels
- #6253 [Engine] macOS / Metal: open contributions, rebased on v0.2.16
- #6281 [Engine] SSD conversation tier turns off when MTP comes from a GGUF (fingerprint looks for mtp/rt/dense.bin)
- #5949 [Engine] HIP/Linux: make HSA_USERPTR_FOR_PAGED_MEM=0 the default on discrete AMD cards
- #6122 [Other] Auto drafting: a PC with several cards reports no card memory, so no default tier can apply
- #5939 [Engine] Windows CUDA12 0.1.59: Tesla V100 GPU is lost during single- and dual-GPU inference
- #6530 [Engine] Hy4-Flash SQ-64 (strata-hy4 0.1.64): long reasoning falls into an endless repetition loop (2 of 2 attempts)
- #6533 [Engine] A branching follow-up re-reads the whole prompt when the final turn is longer than 1024 tokens
- #6531 [Benchmark] gfx1100 (RX 7900 XTX, IQ3_S): existing switches give 1.9–3.7x prompt and 1.2x decode speedup
- #5893 [Engine] API key changes when unloading and reloading a model, invalidating existing API clients
- #6521 [Benchmark] R9700 + Threadripper PRO 3975WX: Hy4-Flash at ~30 tok/s (SQ-64, 4K) vs. 62 tok/s in the 0.2.18 notes
- #6522 [Engine] Ollama-converted GGUF fails at load: qwen35moe.attention.head_count_kv is a per-layer array
- #6523 [Engine] Windows NUMA in Strata: Why Our 40-Core Server Initially Used Only 20 CPU Workers
- #6526 [Engine] gfx1200 (2x RX 9060 XT, ROCm 10.1): peer-GPU fused WMMA prefill, RDNA4 MMVQ load_int_b2/Q6KTraits, and hipBLASLt fallback notes
- #6525 [Engine] Drop in MTP head replacement trained against quantized model output: +3.7 / +4.4 % decode at --spec 3
- #6339 [Engine] 0.1.62: --lookup-max-t past 8 with --draft dflash2 ends the request (dflash2: draft rows out of range); a small fix, +11..13 % on file rewrites
- #5614 [Engine] RX 9070 XT gfx1201: 0.1.58 verify unspecified launch failure triggers MES REMOVE_QUEUE GPU reset
- #6518 [GUI] Model picker shows "Download size" for a fully downloaded Qwen-family model (IQ3_S); Gyro-S in the same folder shows "Already on this computer"
- #6435 [Benchmark] RX 9070 (gfx1201, Windows 11, 32 GB RAM) / Gyro-S TQ1_0 / 0.2.6 → 0.2.17: 0.2.17 defaults lose ~30% prompt and ~12% decode on a 32 GB PC; two settings bring it back
- #6517 [Other] Feature request: rename-aware editing, an integrated debugger, and an optional reviewer model
- #5629 [Engine] GPU fatal error: "layer X never rang (an illegal memory access was encountered)"
- #6511 [Benchmark] 2x Tesla P40 (sm_61) split: 0.2.17 gives +9% decode, but 220K prefill is still ~5% under the 0.2.13.1 MMQ path (same-day A/B)
- #6431 [Engine] 0.1.63 STRATA_L2_PF=1 default regresses NVFP4 batch decode by 8-12% on 2x RTX 5090 with PCIe-streamed experts
- #6479 [Engine] CPU/GPU SwiGLU discrepancy from exponential evaluation on NVIDIA Ampere
- #6515 [Engine] gfx12: `nw_barrier()` falls back to `__syncthreads()`, so the fused prompt experts' stage writes can race (latent)
- #6513 [Engine] Linux HIP: engine crashes at start (SIGSEGV in libamdhip64) with CachyOS's hip-runtime-amd 10.1 v3/znver4 builds; Arch's build works
- #6489 [Engine] 16K cold prefill regression on a single RTX 4090 D (sm_89, IQ3_S): 0.1.55 → 0.1.60/0.1.63 (−21 %), unrelated to --prefill, --expert-cache or the route tail skip
- #6273 [Engine] 16 GB cards: guard the ~1 GB VRAM floor during KV growth and document the per-format context limit
- #6232 [Engine] 384K compaction re-reads 264K tokens (456 s): content-addressed, position-independent segment cache for near-zero-reread compaction
- #6509 [Engine] Engine 0.1.55+ decodes 35–55% slower than 0.1.52 on RTX 3080 Ti + RTX 4070 (Windows): decode expert-cache hit falls from ~92% to 55–70% although the cache gets ~40% more slots
- #6507 [Engine] q36: one early byte change costs the whole prompt — all-or-nothing match_len has no GDN state to step back to (0 reused + 38026 read, 108 s)
- #6074 [Engine] 0.1.60 prompt kernels: -13% at 20,000 tokens on a 16 GB card, +8% at 40,000, flat at 16,000
- #6483 [Engine] Long Gyro-S decode triggers HIP 719 / LiveKernelEvent 141 on RX 9070 XT
- #6503 [Engine] Swift IQ3_S_G8 (GDN Q8_0) returns garbled token soup with engine 0.1.63 on sm_120, while the same files decode correctly in another engine
- #5839 [Engine] 0.2.13 quoted EOS still ends agent turns with reasoning-only stop, with opt-in recovery results
- #6500 [Engine] Vulkan backend: scope for the port (stage 1 landed: the loop and one Q4_0 kernel)
- #6497 [TUI] Strata uses more threads that settings says
- #6327 [Engine] Compaction re-reads 331K-342K tokens five times (19.5-20.1 min each): the cache budget is the blocker (PR #4338)
- #6496 [Other] Windows HIP rdna2 package (0.1.63) works on a real RX 6900 XT (gfx1030)
- #6495 [Engine] Layer-major streaming prefill on Metal
- #6493 [Engine] serve loop: GENI prompt can write past the position table; --cvec offsets unchecked (static analysis)
- #6491 [Other] Ray, an assistant and coding agent built on Strata: would any part of it help the app?
- #5787 [Engine] serve: python.exe dies natively under long agent sessions — every dump faults inside CPython on the same nested `any(genexpr)` path (`serve/frontend.py` `_has_tag`)
- #5751 [Engine] Model load hard-powers-off the whole PC on RTX 5070 Ti (Kernel-Power 41, no dump): autotune variants and 0.1.58 crash; 0.1.46 + conservative VRAM settings stable
- #6487 [Engine] 0.1.62: --vision cpu fails deterministically at startup with "kv_stream: reset: an illegal memory access" (sm_86, Swift IQ3_S) — --vision gpu on the same box is fine
- #6409 [Engine] Qwen3.8-Next-Flash on Strix Halo AI Max 395 (128GB UMA): tuning for 2 concurrent streams at >100 tok/s each @ 200k context
- #6478 [Engine] Expert-cache hit rate and decode throughput regress from engine 0.1.52 to 0.1.62/0.1.63 on two 11 GB Turing cards (Qwen3.8-Flash-Next GSQ-RCO IQ3_XXS)
- #6472 [Benchmark] RTX 4090 24 GB VRAM headroom for long-context KV and IQ4 experts
- #6404 [Engine] gfx1030 (RX 6800 XT, Windows): 0.1.63 reads prompts 2x faster than 0.1.60 but decodes 11 % slower — same config, A/B on one host
- #6470 [Engine] CUDA Q8_0 mmproj loses its VRAM saving with the native picture reader
- #6469 [GUI] The desktop agent's compaction request re-reads the conversation from the system prompt on Gyro (hybrid); --prompt-cache-spread cuts it from 14.9 to 2.7 s here
- #5672 [Engine] gfx1100 (RX 7900 XTX) --adapt-async 1 results in timeout followed by a display-driver reset during long generations
- #5943 [Engine] --prefill auto keeps a 5250-slot lend region on a fully-resident 3x 32 GB split (0.1.58 kept 0): decode -26%, prefill -46%; STRATA_PREFILL_LEND_PCT=0 restores the full tier
- #6466 [Engine] 0.2.14 was faster than 0.2.17 rx 9060 xt (engine 0.1.61 vs 0.1.63)
- #5953 [Engine] Qwen3.6-35B-A3B fails at startup on gfx1150 (Linux, ROCm 7.15): prefill gemm cublasGemmEx f16 - cuBLAS status 6
- #6429 [Benchmark] hipBLASLt tuning table by default on gfx1100 (RX 7900 XTX, IQ3_S): speed, decode and logits
- #6467 [Engine] 0.2.17 on Windows: the CPU picture-reader fallback dies with "Bad file descriptor" (the encoder log fd is read as a path)
- #6463 [Engine] Claude Code's small side requests evict the agent's parked conversation (default conversation cache)
- #6462 [Benchmark] RTX 4090 (sm_89): sm_86 bit-exact set +6%, sm_120 switches +4% on UD-Q4_K_XL, CPU share 32/6144 cuts 3K-token follow-up turns 26%
- #6461 [Engine] A tool parameter named `function` is dropped even when the tool declares it (Playwright MCP `browser_evaluate`), patch included
- #6460 [Engine] Claude Code auto-compaction returns an empty summary near the context limit unless `anthropic_thinking` is `on_request`
- #6459 [Engine] Codex CLI auto-compaction fails on /v1/responses: `compaction_trigger` input items are rejected (400)
- #6448 [Engine] Scope: plan an integrated Intel GPU (Arc B390, Panther Lake) as one pool of memory with the system in the SYCL engine
- #5678 [Engine] Qwen3.8-Flash-Next IQ3_S on sm_121 (unified memory): prefill stalls at 33,372 of 33,379 tokens, just past the 32,768 chunk; STOP not honoured, restart never finishes
- #5990 [Engine] V100 (sm_70) on Linux: 13× Xid 79 + PCIe CmpltTO across engines 0.1.35–0.1.59.1 — not version-specific
- #6121 [Engine] Non-finite logits (#879) with Unsloth UD-Q4_K_XL native on RTX 4090 — deterministic at 267 tokens; IQ4_XS and Q5_K_XL clean on same engine
- #6208 [Engine] 0.2.15 decodes 28-36% slower than 0.2.7 on RTX 3060 12 GB: --expert-cache auto starts with a much worse resident set (same pack and profile)
- #6445 [Engine] Abort on UR_RESULT_ERROR_DEVICE_LOST at the verify window (Intel Arc Pro B70, Linux)
- #6077 [Engine] PLE: a VRAM hot tier for n-gram table rows, and stop pinning what is already resident
- #6440 [Engine] Release duplicate dense-loader weights on resident Metal
- #5995 [Engine] Exit code 0xC0000409: stack buffer overrun on V100 + RTX A4500
- #6118 [Engine] Cold 50K prefill regression: 0.1.52 → 0.1.59 → 0.1.60 on RTX 3060 12GB (−16.5%)
- #6437 [Engine] Replies end mid-word with finish_reason "stop" at ~140K context (int8 KV, MTP on)
- #6436 [Engine] Park the engine: give all VRAM back between requests, keep the process, the context and the conversation
Pull requests
- #4310 [Engine] session files: format v2 with per-stage save/restore under --layer-split
- #6520 [Engine] Inkling-2-Mini from the official NVFP4/MXFP8 checkpoint, expert offload and CPU experts for 8-16 GB cards
- #6210 [Engine] Opt-in P2P q8_1 decode push, fused P2P down-projection, and GPU peer sync flag
- #6529 [Engine] Volta sm_70 decode optimizations, zero-VRAM in-place Q8_0, NVLink split head, and volta-forever detection
- #6548 [Engine] Add opt-in streamed-KV QSA slot-group batching, pre75 multi-slot attention, and batch SIDE_MIXER
- #6451 [TUI] strata launch --model-path: name Swift's shard 1 as --ple-gguf, where its n-gram table is
- #6547 [Engine] Add opt-in multi-slot batched MTP drafting (STRATA_BATCH_MTP_FUSED)
- #6546 [Engine] Opt-in resize_for_overwrite and parallel host memcpy for kv_mode != 0 park/restore
- #6545 [Engine] Size STRATA_PARK_ASYNC VRAM stage by device-resident bytes under kv_mode != 0
- #6544 [Engine] Fix O(N^2) scan in getline_fd and avoid full-line copy in BGEN/parse_i64_list
- #6543 [Engine] Guard negative page_table entries in qsa_prompt_attn and track monotonic kv_stream overflow_count
- #6524 [Engine] SYCL: enable the conversation cache's SSD tier (0.2.17's ConversationSpill)
- #6439 [Engine] Persist the SSD conversation tier across restarts when the model fingerprint matches
- #6305 [Engine] fingerprint the resolved MTP files, so the SSD conversation tier stays on for GGUF and rt drafters
- #6539 [Engine] Hy4-Flash (Metal): resume at the longest common prefix, several conversation slots, SSD tier kept across restarts
- #6538 [Engine] SSD conversation tier: fingerprint the resolved MTP files and keep the tier across restarts (v0.2.18)
- #6534 [Engine] sycl: keep the branch point a follow-up can reach
- #4306 [Other] serve: session ladder — conversations survive an engine unload
- #3551 [Engine] add an opt-in CUDA and HIP hybrid build
- #4551 [Engine] session-file v3: incremental append saves under --session-save-append
- #3773 [Engine] q36: MTP drafts for sampled requests behind STRATA_Q36_MTP_SAMPLED (stacked on #3678)
- #6535 [Benchmark] RTX 5060 Ti / Qwen3.8-Flash-Next / IQ3_S / strata-qwen38-thinking-v1-draft.1
- #6532 [Engine] prefill: opt-in STRATA_I8_DEFER - unpicked streamed entries no longer cut the int8 expert groups into fragments (gfx1030 4K prompt +3.6%)
- #5897 [Engine] serve: let the "vision" section carry its own "env" for the image encoder
- #3800 [Engine] Token mask for JSON response formats: MTP drafts survive constrained decoding
- #5436 [Engine] opt-in STRATA_ARENA_NUMA: place the expert arena per NUMA node on two-socket Linux
- #6506 [Engine] add Inkling 2 model family and dual-GPU looped MoE engine support
- #6399 [Engine] Prefetch routed expert weight rows into L1 on GB10
- #3607 [Engine] native Windows support for Intel Arc SYCL, rebased onto 0.2.18 (engine 0.1.64)
- #6365 [Engine] Prefer FP32 TLA sparse prefill on Battlemage in builds with local TLA headers
- #6514 [Benchmark] RX 7600 XT (gfx1102): ROCm 7.2.4 vs 10.1, and the CachyOS ROCm 10.1 crash
- #4928 [Engine] long --lookup-max-t windows over a layer split (opt-in, single conversation)
- #6510 [Engine] --experts-wex4: read the routed experts from a checked WEX4 container (opt-in, file tier)
- #6102 [Engine] use runtime expert count in layer-major prefill
- #4376 [Engine] dsv41: the WMMA expert kernels compile on pre-Volta cards
- #4366 [Engine] DeepSeek prompt: the busy experts take FP16 GEMMs, not the tiles (DS4_FAT_ROWS, opt-in)
- #6197 [Engine] int8 expert kernels: IQ4_XS as a down format
- #6176 [Engine] q4u32: the 4-bit KV cache with a fitted step as the default on RDNA4
- #6498 [Benchmark] RTX 3060 12GB / Qwen3.8 Flash-Next Q2_0 / standard v1 draft.1 + 260K xhigh
- #6492 [Other] tests: the first VisionTempFiles class was shadowed and never ran
- #6490 [Engine] Load pack weights directly into shared Metal storage
- #6419 [Engine] autonomous hardware-synchronized dual-GPU decode window (P2P zero-wait) [Draft speed Lab 230 tok/s+ RFC]
- #6485 [Engine] Add experimental gfx90a HIP build
- #2965 setup: a separate opt-in layers 8-33 projection vector for the original Qwen3.8-Flash-Next (any quantization)
- #6477 [Benchmark] Arc Pro B60 / Qwen3.8-Flash-Next / IQ3_S / strata-qwen38-thinking-v1-draft.1
- #3957 [Engine] serve: a long conversation survives an engine restart - parked conversations saved at a graceful stop and loaded at the next start, to files or (Windows) kept in RAM with no disk writes (opt-in STRATA_PARK_DIR, STRATA_PARK_RAMLIKE=1; layer split included)
- #5476 [Engine] gfx103x decode: opt-in one-warp MMVQ for K <= 4096 and a per-format expert layout table (STRATA_MMVQ_NW1, STRATA_EXP_TABLE)
- #5475 [Engine] opt-in int8 copies for the persistent hc read (STRATA_HC_PERSIST_Q8, STRATA_HC_I8ONLY)
- #2843 [Engine] Slot checkpoints for clients that rewrite the end of their history (3/3, depends on #2527 and #2528): STRATA_SLOT_CHECKPOINTS, STRATA_BRANCH_CHECKPOINT, BCHECKS
- #2528 [Engine] Slot placement policy for STRATA_SLOT_CACHE_AFFINITY (2/3, depends on #2527): protection by history length, retention scores, value decay, no wait
- #2527 [Engine] Keep conversations in pipelined batch slots (1/3): active-row windows, park=, cache-aware placement behind a policy interface
- #1614 [Engine] shared KV pool for all lanes (--kv-pool-tokens): at 2 x 256k the first token comes 2-5x sooner and lane moves ~10x faster; off = identical to main
- #6464 [Benchmark] Strix Halo 64 GB unified memory (Ryzen AI Max+ 395)
- #6453 [Engine] SYCL: an integrated Intel GPU is planned as one pool of memory with the system (Arc B390)
- #6452 [Engine] serve: an on-demand CPU picture reader starts without asking the expert cache for VRAM
- #6385 [GUI] preserve completed summaries on context retry
- #6390 [GUI] restore the local agent reply ceiling at large context sizes
- #6391 [Other] verify the context patch on reused desktop Harness stages
- #6394 [GUI] preserve long-running desktop agent sessions
- #6458 [Engine] STRATA_FILL_STAGED: copy mapped blobs to the GPU via pinned buffers
- #6455 [Engine] SYCL: an Xe3 GPU (Panther Lake, Arc B390) runs the Battlemage kernels; STRATA_INTEL_GEN sets the class
- #6454 [Engine] serve: an integrated Intel GPU's memory is the system RAM in the Monitor and the free-memory checks
- #3919 [Engine] kv: size the resident ring from the run's depth (opt-in, STRATA_KV_RING_ADAPT)
- #3729 [Engine] SYCL: flag Intel iGPUs as integrated so the expert cache is sized from MemAvailable
- #3648 [Engine] --peer-device host route: STRATA_PF_PEER_PIECES=1 (opt-in) moves the activations and the sums in 8 pieces (+7 % prompt on 2x V100)
- #6447 [Engine] Size Metal state after batch admission
- #6446 [Engine] Release Metal streaming resources when the window closes
- #6443 [Engine] Avoid redundant top-K scans in Metal expert read-ahead
- #6434 [Engine] Use eight-row tails in M1 resident expert GEMMs
- #6432 [Engine] Reuse Metal resident prefill scratch buffers
- #6441 [Engine] Reduce resident Metal QSA prefill scratch
- #2427 [Engine] HC_Q8: skip the layers' bf16 hyper-connection copies on a layer split too
- #6427 [Engine] layer split auto with DFlash2: plan DFlash2's drafter and after-cache bind on the last stage, and keep the gate-off fallback loadable
- #4766 [Engine] gfx906: build dflash2 for Radeon VII
- #6430 [Engine] q36: --kv fp32|fp16|fp8|fp4 for the K/V cache — 262,144 tokens on one 12 GiB card