z3st
5bb32e6bdb
feat: open architecture filter, expand search keywords, let platform decide compatibility
2026-07-24 19:58:45 +08:00
z3st
75ba3bb301
chore: switch GPU to Iluvatar_bi-100
2026-07-24 19:49:42 +08:00
z3st
e28f2e3eca
feat: switch to llama.cpp with GGUF support on Iluvatar, keep only GPTQ/AWQ filtered
2026-07-24 19:29:48 +08:00
z3st
e0b6c3b5a8
feat: use file-based SQLite database for persistent failure tracking
2026-07-24 19:03:13 +08:00
z3st
f438206c0e
feat: submit only 2 models, add failed model tracking to avoid re-submission
2026-07-24 18:57:52 +08:00
z3st
12616c3816
restore submit version from v2.1.0
2026-07-24 18:56:18 +08:00
z3st
0a4ba2e146
feat: standalone cancel-all version - startup auto cancels all waiting/running tasks
2026-07-24 18:53:24 +08:00
z3st
2222e1545b
feat: cancel-all mode - cancel all waiting/running tasks on startup
2026-07-24 18:51:51 +08:00
z3st
441b540e47
feat: filter out GPTQ/AWQ formats and Qwen3.5 arch (incompatible with Iluvatar)
2026-07-24 18:50:21 +08:00
z3st
afbda884ad
feat: multi-page search, filter downloads 50-5000 for cold models
2026-07-24 00:10:32 +08:00
z3st
68c7a99e68
fix: replace gpu variable with TARGET_GPU in submit record
2026-07-24 00:02:35 +08:00
z3st
8e8fd50927
chore: switch target GPU to Iluvatar_bi-150
2026-07-23 23:51:28 +08:00
z3st
5a6862f8da
fix: revert to ModelScope search (HF unreachable in container), fix ACCOUNTS reference
2026-07-23 23:32:33 +08:00
z3st
727f0f678f
feat: single GPU, HuggingFace only, vllm only
...
- Target single GPU (ppu_zw_810e) instead of 4 GPUs
- Search only from HuggingFace API (removed ModelScope)
- Framework fixed to vllm (removed llama.cpp)
- Filter out GGUF format models
- Simplified config and functions
2026-07-23 23:19:20 +08:00
z3st
207d44b8f2
fix: add nv_framework back for API safety
2026-07-23 22:43:22 +08:00
z3st
1d33394dac
fix: remove nv_framework, let platform auto-assign
2026-07-23 22:42:45 +08:00
z3st
1811a66aee
fix: match platform auto-generated configParams format
...
- Remove /bin/bash -ic wrapper from sut_config.command
- Use list format for all command arguments (matches platform auto-gen)
- Add nv_framework field
- Remove unnecessary fields: docker_image, nv_docker_image, modelFormat, model, cpu_num
- Fix ref_config to match platform format (port 80, max_model_len 4096)
2026-07-23 22:35:50 +08:00
z3st
12bf6d637f
fix: move gpu_num/cpu_num outside values block to match API spec
2026-07-23 22:10:17 +08:00
z3st
85c456afe0
fix: yaml.dump width=1000 to prevent command string wrapping
2026-07-22 22:43:35 +08:00
z3st
b3b52848c2
fix: verifyResult can be None, ensure dict return
2026-07-22 01:39:33 +08:00
z3st
d6e9f2e7a0
fix: page_size limit 50 for ModelScope API
2026-07-22 01:13:26 +08:00
z3st
821edbc8f5
debug: add logging to search function
2026-07-22 01:08:03 +08:00
z3st
5e2df178e1
feat: re-enable auto-run pipeline after connectivity test
2026-07-21 20:45:26 +08:00
z3st
761e74ff82
feat: auto-run connectivity test on startup, print results to logs
2026-07-21 20:31:23 +08:00
z3st
09ac80f1ec
feat: add /test endpoint for connectivity check, disable auto-run
2026-07-21 19:53:28 +08:00
z3st
3975a3370a
fix: correct ModelScope API endpoint and field names
2026-07-21 19:31:35 +08:00
z3st
7313f8da67
feat: auto-run pipeline on startup
2026-07-21 19:09:41 +08:00
z3st
2955a57e2e
feat: full cloud pipeline - search, filter, submit
2026-07-21 19:04:43 +08:00
z3st
fefc5688e1
init: minimal submit agent
2026-07-21 18:45:22 +08:00
