Zuraiz Anjum e7c2edb857 serve: honor stop and stop_sequences, cut the answer before the stop string and stop the engine there

2026-10-05 13:16:53 +02:00

..

benchmarks

AMD HIP backend for gfx1100 (PR #121, squashed onto its 0.1.21 base)

2026-09-29 14:33:45 +02:00

media

README: How does it work? in plain words, with illustrations in the app's style (colors, icons, font; light and dark)

2026-09-28 14:38:40 +02:00

paper

Strata: Qwen3.8-Flash-Next on a 12-24 GB NVIDIA GPU + 64 GB RAM, one-click install

2026-09-24 18:42:20 +02:00

AI_SETUP.md

setup: Unsloth's UD-IQ4_XS is a regular choice (no longer experimental); UD-Q4_K_XL stays experimental

2026-10-04 13:24:29 +02:00

AMD_HIP_PERFORMANCE.md

Low-RAM mode, resident variant: the experts the GPU does not hold stay in RAM (--resident-experts), chosen by setup

2026-09-30 16:51:04 +02:00

AMD_HIP.md

docs: Older GPUs (experimental) - the support matrix, the CUDA 12 engine and how to opt in

2026-10-03 23:44:14 +02:00

BATCHING.md

#465 setup: recommend "parallel" only where the experts mostly fit in VRAM

2026-10-04 03:36:22 +02:00

COMMUNITY_BENCHMARKS.md

docs: the host has 64 GB of RAM (62 GiB usable), not 62 GB

2026-10-05 12:18:48 +02:00

DETAILS.md

serve: honor stop and stop_sequences, cut the answer before the stop string and stop the engine there

2026-10-05 13:16:53 +02:00

HOW_IT_WORKS.md

docs: a short README for regular people; the details move to INSTALL, MODELS, TROUBLESHOOTING, HOW_IT_WORKS

2026-10-01 20:08:58 +02:00

INSTALL.md

docs: Older CPUs (experimental) - what setup does on a CPU without AVX2, what was measured

2026-10-04 02:13:51 +02:00

INTEL_ARC.md

docs: INTEL_ARC says what ran on an Arc (#868 follow-up)

2026-10-05 12:18:48 +02:00

INTEL.md

sycl: the stream-all prompt walk by default when the VRAM holds >90% of the experts

2026-10-03 16:32:22 -05:00

MCP_SERVER.md

setup: Unsloth's UD-IQ4_XS is a regular choice (no longer experimental); UD-Q4_K_XL stays experimental

2026-10-04 13:24:29 +02:00

MODELS.md

setup: Unsloth's UD-IQ4_XS is a regular choice (no longer experimental); UD-Q4_K_XL stays experimental

2026-10-04 13:24:29 +02:00

MULTI_GPU.md

setup: the low-RAM mode's resident variant on several GPUs (#642)

2026-10-05 11:03:50 +02:00

NVIDIA_V100.md

docs: Older GPUs (experimental) - the support matrix, the CUDA 12 engine and how to opt in

2026-10-03 23:44:14 +02:00

OLDER_GPUS.md

setup: the CUDA 12 engine's driver floor per OS (Windows 527.41 -> 528, Linux 525); the wheel pins match the 12.9.1 toolkit

2026-10-04 08:01:18 +02:00

ORCA_Q4_K_S.md

#296 follow-up: only the Q5_0 MMQ instance and case stay (Q4_K / Q5_1 come with S5's STRATA_MMQ_KQUANTS)

2026-10-01 10:31:23 +02:00

ORCA.md

Support Orca IQ3_XXS with explicit BF16 compatibility packing

2026-09-27 14:10:52 -06:00

SECOND_GPU.md

setup: --remote-expert-opt recommended for 2+ GPU configs (#578)

2026-10-04 07:57:23 +02:00

TROUBLESHOOTING.md

setup: STRATA_NVCC picks the CUDA toolkit; CUDA_HOME is looked at; glibc 2.43 + CUDA 12.9 and gcc 15 documented (#601)

2026-10-03 18:25:30 +02:00

UNSLOTH_Q4.md

native experts: Q4_0 and Q4_1 GPU kernels, Q4_0 PLE table

2026-10-05 12:18:48 +02:00