Basho
Basho is Murakumo's writing face. It is an inference organism, not a new standalone checkpoint: a draft member produces an answer and a review member accepts or corrects it under Basho's writing profile.
| Field | Value |
|---|---|
| Public API model ID | ao/basho |
| Endpoint | https://api.murakumo.cloud/v1/chat/completions |
| Current draft member | murakumo-main |
| Current review member | ling-3.0-tiny |
| Dedicated metered alias | awai-network/basho (identified callers; currently resolves to zai-org/GLM-5.3-Flash) |
| Dedicated runtime source | tacos4me/GLM-5.3-Flash-NVFP4-FP8ATTN-512K at fd5efd8c534f6170fa9b4143ec57e0eed6faf842 |
| Weights in this repository | None; member and dedicated checkpoints remain separately licensed artifacts |
curl https://api.murakumo.cloud/v1/chat/completions \
-H 'content-type: application/json' \
-d '{
"model": "ao/basho",
"messages": [{"role": "user", "content": "Rewrite this paragraph clearly: ..."}],
"max_tokens": 512
}'
The public name is stable while internal membership may change. Clients should
therefore select ao/basho, not a current member checkpoint. A live probe on
2026-09-22 returned HTTP 200.
Dedicated serving runtime
This repository also contains the matched vLLM runtime used to expose a
dedicated checkpoint as awai-network/basho: a pinned container definition,
GLM5 Next block-FP8 attention and language-head loader patches, and measured
SM120 kernel configurations. entrypoint.sh pins both the external checkpoint
repository and its revision. The checkpoint weights and base container are not
copied into this repository.
Behavior
Basho checks requested length, format, and factual dependencies before answering. Its reviewer changes only claims or structure that fail those conditions, preserving the draft's voice where possible. Generated text still requires normal human review for consequential use.