Model← Models
BTL-4 Compact (IQ2_XXS)
35.1B mixture of experts, 2.1B active, IQ2_XXS (2.30 bpw), 10.0 GB.
- Parameters
- 35.1B / 2.1B active
- Quant
- IQ2_XXS (2.30 bpw)
- Size
- 10.0 GB
- Family
- btl
btl-4
RejectedMoEresidentllama.cpp
Model card ↗Outcome as of 20 September 2026, from the seat registry. The write-up below was made when the model was tested.
Why it was tested
The only build badtheorylabs publish. Tested to see what an aggressive IQ codebook costs on a sparse MoE.
What came of it
Rejected in favour of the Q4_K_M build. Same weights, and the quantisation alone cost 27 points of worldview and 17 of toolcall, plus 2.06x throughput (23.6 vs 48.5 tok/s). Evidence for iq-codebook-penalty-transfers-to-moe-kquant-does-not.
Fine-tuned from BTL-4 (Q4_K_M).
| Suite | Score | 95% interval | Passed | Runtime · profile | Recorded |
|---|---|---|---|---|---|
| chainwalk | 0.0% | 0.0–65.8 | Passed 0/2 | llama.cppagentic-fast | 2026-09-15 |
| contract | 64.6% | 50.4–76.6 | Passed 31/48 | llama.cppextraction | 2026-09-15 |
| enrichment | 87.0% | 75.6–93.6 | Passed 47/54 | llama.cppextraction | 2026-09-15 |
| epistemics | 100.0% | 92.6–100.0 | Passed 48/48 | llama.cppstudy-nothink | 2026-09-15 |
| grounding | 76.8% | 68.7–83.3 | Passed 96/125 | llama.cppextraction | 2026-09-15 |
| nullschema | 36.1% | 26.0–47.6 | Passed 26/72 | llama.cppextraction | 2026-09-15 |
| screening | 95.0% | 86.3–98.3 | Passed 57/60 | llama.cppextraction | 2026-09-15 |
| synthesis | 80.0% | 68.2–88.2 | Passed 48/60 | llama.cppextraction | 2026-09-15 |
| tagger | 85.0% | 79.9–89.0 | Passed 204/240 | llama.cppextraction | 2026-09-15 |
| tagger_nullable | 91.7% | 87.5–94.5 | Passed 220/240 | llama.cppextraction | 2026-09-15 |
| toolcall | 76.7% | 64.6–85.6 | Passed 46/60 | llama.cppextraction | 2026-09-15 |
| tools | 100.0% | 79.6–100.0 | Passed 15/15 | llama.cppagentic-fast | 2026-09-15 |
| toolscale | 100.0% | 91.2–100.0 | Passed 40/40 | llama.cppagentic-fast | 2026-09-15 |
| worldview | 62.5% | 47.0–75.8 | Passed 25/40 | llama.cppextraction | 2026-09-15 |
Runs marked unenforced were measured through mlx_lm on a suite that requires constrained decoding. mlx_lm accepts the schema and ignores it, so those numbers do not measure what the suite is designed to measure. They are kept and labelled rather than deleted.
Every run log entry for this model (14)
| Finished | Suite | Runtime · profile | Score | Coverage | Grader · dataset · config |
|---|---|---|---|---|---|
| tagger_nullabletagger-nullable/1 | not recordedextraction | 91.7% | Coverage 240/240 | g 04e875431e26c4b5d db3addf23a2bc4cec f9c64867e9dfb91b | |
| taggertagger/2 | not recordedextraction | 85.0% | Coverage 240/240 | g 63e04d139a0e5ed3d 7ef587bf33e5788ac 81d3e449bf19eaea | |
| enrichmentenrichment/3 | not recordedextraction | 87.0% | Coverage 54/54 | g b51222c2ca38cfdfd 75361864baa727abc 0b71856ba217c91c | |
| contractcontract/2 | not recordedextraction | 64.6% | Coverage 48/48 | g cc2d23c5bae45474d 1fe53d30cb5480f6c bf1f8546a0e1b97c | |
| chainwalkchainwalk/3 | not recordedagentic-fast | 0.0% | Coverage 2/2 | g f4bfcf6f040b0978d 129a9e74408c8919c 8dbbe23a8427e3bc | |
| epistemicsepistemics/4 | not recordedstudy-nothink | 100.0% | Coverage 48/48 | g acf84972a0ad5f38d 4bce453fd376553cc 438aa2f62c6153af | |
| groundinggrounding/1 | not recordedextraction | 76.8% | Coverage 125/125 | g 30b933dcc6a7e2d4d dcb396faf8b83b0ac 7d9f8293fd874d62 | |
| screeningscreening/1 | not recordedextraction | 95.0% | Coverage 60/60 | g 0661b49a10be2d2ad 64c97ca2269c29afc 8197a1913a41df95 | |
| synthesissynthesis/2 | not recordedextraction | 80.0% | Coverage 60/60 | g 4156dd3c13361e8bd 267d1954e225c5b6c 11c6d6217d23f0b7 | |
| worldviewworldview/2 | not recordedextraction | 62.5% | Coverage 40/40 | g 9acfba1b704feb28d e97835de9e6fc227c 0104404569f7c535 | |
| nullschemanullschema/1 | not recordedextraction | 36.1% | Coverage 72/72 | g d7448e0f5fc3f9afd 3b91280bb40372efc 9b2dcb073c569fce | |
| toolstools/1 | not recordedagentic-fast | 100.0% | Coverage 15/15 | g 2763675a5c3da564d 33a44812f0ad4c39c e43140ad3e77d008 | |
| toolscaletoolscale/1 | not recordedagentic-fast | 100.0% | Coverage 40/40 | g a2bc3751abfa28e0d b24abb108c409178c 962089edc90681de | |
| toolcalltoolcall/1 | not recordedextraction | 76.7% | Coverage 60/60 | g 804e8fc9457649d8d 10af88c170513155c 7d3d7741437436e1 |
- BTL-4 (Q4_K_M)Measured, not seated
35.1B · MoE · Q4_K_M (community conversion) · 20.5 GB
Fine-tuned from Ornith-1.0-35B on an execution-gated corpus: trajectories kept only where the code ran and passed. 256 experts, 8 routed, 262k native context, ~20 KB/token KV.