1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
|
schemaVersion: "1"
# Sandbox that serves an Olala GGUF model with llama-server (OpenAI-compatible API)
# on the host, using the OVHai llama.cpp fork (CPU only) from kits/llama-cpp-olala.
#
# sbx env create sbxenv.yaml # create + download + serve
# sbx env create sbxenv.yaml --env-arg host_port=9090 # another host port
# sbx env create sbxenv.yaml --env-arg models_dir=./my-models # another models folder
# sbx env run sbxenv.yaml # attach Claude Code
# sbx env exec --no-input sbxenv.yaml -- bash -lc '$SBX_WS/scripts/olala.sh serve' # restart after `sbx stop`
# sbx env rm sbxenv.yaml # models stay in models_dir
#
# Every step of scripts/olala.sh is skipped when already done (llama-server found,
# model present, server running), so re-running it is cheap.
# Use the same --env-arg values with every sbx env command of a given sandbox.
name: houla-olala
agent: claude
workspace: .
kits:
# llama-server / llama-cli from the OVHai fork (built by sbx from the local kit).
- ./kits/llama-cpp-olala
args:
model_repo:
default: OVHaiLLM/Olala-v1.0-mini-7A1B-GGUF
description: Hugging Face repository of the GGUF file
model_file:
default: Olala-v1.0-mini-7A1B-q8_0-safe2.gguf
description: GGUF file to download and serve (q8_0-safe2 = recommended for CPUs)
models_dir:
default: ./models
description: Where models are stored; relative to the repository root, kept when the sandbox is removed
host_port:
default: "8080"
pattern: "[1-9][0-9]*"
description: Host port that publishes llama-server
threads:
default: "16"
pattern: "[1-9][0-9]*"
description: CPU threads used by llama-server (leave a couple of CPUs free)
ctx:
default: "32768"
pattern: "[1-9][0-9]*"
description: Context size (shared by the parallel slots)
env:
ANTHROPIC_MODEL: claude-opus-5-5
OLALA_MODEL_REPO: "${{ env.args.model_repo }}"
OLALA_MODEL_FILE: "${{ env.args.model_file }}"
OLALA_MODELS_DIR: "${{ env.args.models_dir }}"
OLALA_THREADS: "${{ env.args.threads }}"
OLALA_CTX: "${{ env.args.ctx }}"
OLALA_PORT: "8080"
SBX_WS: "${{ env.fileDir }}"
ports:
- sandbox: 8080
host: ${{ env.args.host_port }}
# Lifecycle commands run on the HOST; they reach into the sandbox with `sbx env exec`.
# postCreate runs once, after creation (not on re-attach).
lifecycle:
postCreate:
# A local sandbox stops as soon as no session keeps it running (and llama-server with it).
# Detached mode keeps it running until `sbx stop`; it is permanent for this sandbox.
- name: Keep the sandbox running in the background (detached)
command: sbx run -d --name "$SBX_SANDBOX_NAME"
timeout: 2m
- name: Download the model (if missing) and start llama-server
# Flags go BEFORE the env file path: after it, sbx reads every word up to `--` as another path.
command: sbx env exec --name "$SBX_SANDBOX_NAME" --no-input "$SBX_ENV_FILE" -- bash -lc "$SBX_WORKSPACE/scripts/olala.sh up"
timeout: 30m
|