Spaces:
Sleeping
Sleeping
Commit History
-
loader: adopt standalone chat_template.jinja; unwrap BatchEncoding in format_chat
Verified signature3cb6615 -
Serve SurjoLabs/Surjo-50m-SFT-Only (int8, 2T) on 7860
Verified signature92dcc4e
-
deploy c66f5aa: decode push
421b8c2
-
Space: AVX2 int8 restored as default (zmm opt-in); log kernel variant
9c7cdba -
Space: AVX512 kernels (zmm fp32/int8, parallel VNNI-Q8) + 2T cap + NEON int4 fix
Sync current engine. Scope: Surjo-50m only, threads [1,2] (2 vCPU box; os.cpu_count sees 16 host CPUs), spec 0/2/4/8, no skip (fresh numbers). Validates zmm dispatch via PPL gate; int8-Q8 path engages only in int8-act mode (separate run).40d0cd2 -
Fix Space bench for real: layout-agnostic blob check (hub 1.x decoupled store)
54a7349 -
Fix Space bench: accept sharded HF blob store (hub>=1.x)
9cf357b -
Restore Space README frontmatter (sdk: docker config)
4dff312 -
Fresh Surjo-50m bench on current code (all 4 quants) + HW inventory
Sync native/src/scripts (vector-exp kernels, HQ4-32 4-bit, 8-acc int8 dot, thread-sweep incl 6T). Scope this run: SurjoLabs/Surjo-50m only, spec 0/2/4/8, no --skip-existing (replace stale 2026-09-16 numbers). entrypoint dumps HARDWARE.txt (lscpu/cpuinfo/meminfo) and re-uploads STATUS.txt on finish.a9c1762
-
Fleet survives ephemeral storage: upload per model + resume from dataset
8825740 -
Serve first, bench second + space-sized fleet (fix 30-min health timeout)
entrypoint starts http.server on 7860 before the fleet so the Space goes healthy in seconds; STATUS.txt tracks progress; --skip-existing resumes after restarts. CMD trims spec to 0/2/4/8 (K8 within ~2pct of best everywhere measured) and threads auto-cap at CPU count. Fleet script: cpu-count thread cap, CISM_SPEC_KS/CISM_THREAD_SWEEP env, --skip-existing.968c40e -
Fix GCC build: <cstring> + <cpuid.h>/x86 guards
a993d03 -
Fix Space config: restore Docker SDK Dockerfile + frontmatter README
The source-tree copy had clobbered both files (generic remote Dockerfile, CISM README without sdk frontmatter). This restores the Space entrypoint (fleet then serve /results on 7860) and the sdk: docker frontmatter.57d2c8f -
CISM remote autobench: full source + fleet runner, serve results on 7860
Builds the native CPU engine, runs the 4-model full AutoBench matrix (all precisions x threads x spec K + HF baselines), then serves /results. Configure via CISM_MODELS/CISM_OUT_DIR variables; optional HF_TOKEN secret backs results up to a dataset.28a1a01
-
Update Dockerfile
Verified signature8928fb5 -
Update Dockerfile
Verified signaturede55b64 -
Update Dockerfile
Verified signature3460caa -
Update Dockerfile
Verified signature0675fb3 -
Update Dockerfile
Verified signaturef6f0400
-
Update Dockerfile
Verified signature43e6b0f -
Update Dockerfile
Verified signaturef836146 -
Update Dockerfile
Verified signature454f91b -
Update Dockerfile
Verified signature9151501
-
Update Dockerfile
Verified signature7033a98 -
Update Dockerfile
Verified signature0402f09 -
Update Dockerfile
Verified signature7cf8efb -
Create Dockerfile
Verified signature2190dc6 -
initial commit
Verified signaturedf14c8b