toktape — published LLM inference runs and their caveats

toktapethe record, not a recording of it

Every run on this page is the file toktape wrote on somebody's own server: the card, the caveats that say when a figure is not quotable, and a replay, all drawn from those same bytes. One word on the machine already running llama-server records yours.

brew install midagedev/tap/toktape other ways to install GitHub

Narrowed toquant: q4_k ×sessions: 1 ×Clear all

41 runs, newest first · showing the first 30

bloomery · Raon-1 mini preview Q4_K_M + DFlash2 · RTX 3090 · 한국어 긴 답변 (전세·월세) 122 tok/sprefill 273 tok/s

raon-1-mini-preview-Q4_K_M.gguf

bloomery0.3.6 (e8bd14c0-dirty)q4_KP43 · G2048 · cache 0%ctx 256krtx-3090 linux1 stream! 1 caveat

midagedev

2026-10-21

bloomery · Raon-1 mini preview Q4_K_M + DFlash2 · RTX 3090 · 한국어 (전세·월세) 124 tok/sprefill 276 tok/s

raon-1-mini-preview-Q4_K_M.gguf

bloomery0.3.6 (e8bd14c0-dirty)q4_KP23 · G387 · cache 0%ctx 256krtx-3090 linux1 stream! 1 caveat

midagedev

2026-10-21

bloomery · Solar-Open2-250B · RTX 3090 · short chat 26.4 tok/sprefill 80.0 tok/s

solar-open2-250b

bloomery0.3.6 (a0288c5c0715)q4_KP30 · G400 · cache 0%ctx 8koffload partialrtx-3090 rtx-a6000 linux1 stream! 3 caveats

midagedev

2026-10-20

bloomery · Solar-Open2-250B · RTX 3090 · 4.5K prompt 25.7 tok/sprefill 293 tok/s

solar-open2-250b

bloomery0.3.6 (a0288c5c0715)q4_KP4560 · G600 · cache 0%ctx 8koffload partialrtx-3090 rtx-a6000 linux1 stream! 3 caveats

midagedev

2026-10-20

bloomery 0.3.1 e659f4fe, Qwen3.8 UD-Q4_K_XL, 48 GiB MemoryMax, BLOOMERY_RESIDENCY=off (arm D) 41.2 tok/sprefill 124 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

bloomery0.3.1 (e659f4feb1b4)q4_KP33 · G256 · cache 0%ctx 4koffload partialrtx-5090 linux1 stream! 1 caveat

juliankang4 anonymous

2026-10-12

bloomery 0.3.1 e659f4fe, Qwen3.8 UD-Q4_K_XL, 48 GiB MemoryMax, residency unset (arm C) 97.4 tok/sprefill 143 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

bloomery0.3.1 (e659f4feb1b4)q4_KP33 · G256 · cache 0%ctx 4koffload partialrtx-5090 linux1 stream! 3 caveats

juliankang4 anonymous

2026-10-12

bloomery 0.3.1 e659f4fe, Qwen3.8 UD-Q4_K_XL, no memory limit, BLOOMERY_RESIDENCY=off (arm B) 41.4 tok/sprefill 127 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

bloomery0.3.1 (e659f4feb1b4)q4_KP33 · G256 · cache 0%ctx 4koffload partialrtx-5090 linux1 stream! 1 caveat

juliankang4 anonymous

2026-10-12

bloomery 0.3.1 e659f4fe, Qwen3.8 UD-Q4_K_XL, no memory limit, residency unset (arm A) 52.3 tok/sprefill 117 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

bloomery0.3.1 (e659f4feb1b4)q4_KP33 · G256 · cache 0%ctx 4koffload partialrtx-5090 linux1 stream! 2 caveats

juliankang4 anonymous

2026-10-12

Qwen3.8-Flash-Next on a 12 GB RTX 3060, short prompt (bloomery 0.3.1 release candidate) 5.4 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

bloomery0.3.1 (5ae948b013c9)q4_KP33 · G400 · cache 0%ctx 28881offload partialrtx-3060 linux1 stream! 3 caveats

midagedev

2026-10-12

Qwen3.8-Flash-Next on a 12 GB RTX 3060, long prompt (bloomery 0.3.1 release candidate) 4.5 tok/sprefill 84.9 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

bloomery0.3.1 (5ae948b013c9)q4_KP5673 · G505 · cache 0%ctx 28881offload partialrtx-3060 linux1 stream! 2 caveats

midagedev

2026-10-12

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf 31.6 tok/sprefill 954 tok/s

bloomery0.3.0 (8d937828)q4_KP5555 · G447 · cache 0%ctx 192kgb10 linux1 streamprompts@v2 ! 1 caveat

juliankang4 anonymous

2026-10-12

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf 54.0 tok/sprefill 449 tok/s

bloomery0.3.0 (8d937828)q4_KP3903 · G256 · cache 0%ctx 7680offload partialrtx-5090 linux1 stream

anonymous

2026-10-12

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf 46.5 tok/sprefill 124 tok/s

bloomery0.3.0 (8d937828)q4_KP32 · G256 · cache 0%ctx 7680offload partialrtx-5090 linux1 stream! 1 caveat

anonymous

2026-10-12

GLM-5.3-Flash-UD-Q4_K_XL-00001-of-00006.gguf 28.9 tok/sprefill 169 tok/s

bloomery0.2.10 (8e55b8ce)q4_KP2067 · G149 · cache 33% · min 79ctx 16koffload partialrtx-a6000 linux1 stream! 2 caveats

midagedev

2026-10-11

GLM-5.3-Flash-UD-Q4_K_XL-00001-of-00006.gguf 27.9 tok/sprefill 139 tok/s

bloomery0.3.0 (8d937828)q4_KP2067 · G159 · cache 50% · min 84ctx 16koffload partialrtx-a6000 linux1 stream! 3 caveats

midagedev

2026-10-11

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf 93.8 tok/s

bloomery0.2.10 (8e55b8ce)q4_KP33 · G400 · cache 0%ctx 16koffload partialrtx-3090 rtx-a6000 linux1 stream! 1 caveat

midagedev

2026-10-10

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf 97.6 tok/s

bloomery0.2.10 (8e55b8ce)q4_KP33 · G400 · cache 0%ctx 16koffload partialrtx-3090 rtx-a6000 linux1 stream! 1 caveat

midagedev

2026-10-10

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf 103 tok/sprefill 1312 tok/s

bloomery0.2.10 (8e55b8ce)q4_KP538 · G1500 · cache 0%ctx 16koffload partialrtx-3090 rtx-a6000 linux1 stream

midagedev

2026-10-10

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf 77.9 tok/sprefill 1698 tok/s

bloomery0.2.10 (8e55b8ce)q4_KP5673 · G387 · cache 0%ctx 16koffload partialrtx-3090 rtx-a6000 linux1 stream

midagedev

2026-10-10

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf 90.8 tok/s

bloomery0.2.10 (8e55b8ce)q4_KP33 · G400 · cache 0%ctx 16koffload partialrtx-a6000 linux1 stream! 1 caveat

midagedev

2026-10-10

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf 94.1 tok/sprefill 1001 tok/s

bloomery0.2.10 (8e55b8ce)q4_KP538 · G1500 · cache 0%ctx 16koffload partialrtx-a6000 linux1 stream

midagedev

2026-10-10

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf 70.9 tok/sprefill 1577 tok/s

bloomery0.2.10 (8e55b8ce)q4_KP5673 · G469 · cache 0%ctx 16koffload partialrtx-a6000 linux1 stream

midagedev

2026-10-10

bloomery 0.2.6 · Qwen3.8 Flash UD-Q4_K_XL · A6000 + 3090 · 3.8K-token prompt, MTP 83.8 tok/sprefill 1450 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

bloomery0.2.6 (39b13149)q4_KP3847 · G425 · cache 0%ctx 8koffload partialrtx-3090 rtx-a6000 linux1 stream

midagedev

2026-10-07

One RTX 3090 · bloomery 0.2.1 · Qwen3.8 Flash Next UD-Q4_K_XL 111 GB · code review, 1500 tokens 71.9 tok/sprefill 461 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

bloomery0.2.1 (b83f1269)q4_KP538 · G1500 · cache 0%ctx 38koffload partialrtx-3090 linux1 stream! 1 caveat

midagedev

2026-10-06

One RTX 3090 · bloomery 0.2.1 · Qwen3.8 Flash Next UD-Q4_K_XL 111 GB · Korean chat 57.3 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

bloomery0.2.1 (b83f1269)q4_KP33 · G400 · cache 0%ctx 38koffload partialrtx-3090 linux1 stream! 2 caveats

midagedev

2026-10-06

One RTX 3090 · bloomery 0.2.1 · Qwen3.8 Flash Next UD-Q4_K_XL 111 GB · Rust chat 66.5 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

bloomery0.2.1 (b83f1269)q4_KP33 · G400 · cache 0%ctx 38koffload partialrtx-3090 linux1 stream! 1 caveat

midagedev

2026-10-06

bloomery 0.2.1 · Qwen3.8 Flash Next UD-Q4_K_XL · RTX A6000 · code review, 1500 tokens 91.1 tok/sprefill 732 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

bloomery0.2.1 (b83f1269)q4_KP538 · G1500 · cache 0%ctx 24064offload partialrtx-a6000 linux1 stream! 1 caveat

midagedev

2026-10-05

bloomery 0.2.1 · Qwen3.8 Flash Next UD-Q4_K_XL · RTX A6000 · Korean chat 69.3 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

bloomery0.2.1 (b83f1269)q4_KP33 · G400 · cache 0%ctx 24064offload partialrtx-a6000 linux1 stream! 1 caveat

midagedev

2026-10-05

bloomery 0.2.1 · Qwen3.8 Flash Next UD-Q4_K_XL · RTX A6000 · Rust chat 80.5 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

bloomery0.2.1 (b83f1269)q4_KP33 · G400 · cache 0%ctx 24064offload partialrtx-a6000 linux1 stream! 1 caveat

midagedev

2026-10-05

bloomery 0.2.1 · GLM-5.3 Flash UD-Q4_K_XL · A6000 + 3090 (--place bp) · code review, 2000 tokens 32.4 tok/sprefill 222 tok/s

GLM-5.3-Flash-UD-Q4_K_XL-00001-of-00006.gguf

bloomery0.2.1 (b83f1269)q4_KP497 · G1003 · cache 0%ctx 2koffload partialrtx-3090 rtx-a6000 linux1 stream! 2 caveats

midagedev

2026-10-05

Older runs →

Newest first unless you choose another order; there is still no rank column and no score. Every row carries the caveats the card would print, because a result set without them is a leaderboard with the sorting taken out.
toktape publish <run.toktape> puts one here. toktape on GitHub

Page views are counted with Cloudflare Web Analytics: no cookies, no fingerprinting. The toktape binary itself sends nothing anywhere.