2fa2c4ae80eab50831abc847c837629547c1be6d
- Author
- TheEdgeOfRage <git@theedgeofrage.com>
- Committer
- TheEdgeOfRage <git@theedgeofrage.com>
- Date
Message
Update default LLM endpoint
Diff
This diff is truncated to protect this page.
1diff --git a/PLAN.md b/PLAN.md
2index c9285a74073ab313ec36fda1c54f58e0897652c5..035aa5ca3dee0947e4722eecfc9d54bb43d75063 100644
3--- a/PLAN.md
4+++ b/PLAN.md
5@@ -1,11 +1,13 @@
6 # Japanese learning RPG — implementation plan
7
8 ## Outcome
9+
10 Create a lightweight local Japanese conversation game in Go. The player moves through a small ASCII city in a Bubble Tea TUI, speaks to workers through a microphone, hears Japanese responses, and gets separate fluency feedback.
11
12 The first demo has three free-roam locations: a ramen shop, a station, and tourist information. There is no save, XP, unlock system, or model-server lifecycle management.
13
14 ## Fixed product rules
15+
16 - UI text is romaji only. Never display kana or kanji in the TUI.
17 - NPC speech plays immediately, but its text stays hidden. `R` reveals romaji; `T` then reveals the already-generated English translation.
18 - Each player turn gets a separate score and short correction in romaji.
19@@ -14,24 +16,27 @@ The first demo has three free-roam locations: a ramen shop, a station, and touri
20 - In-game prompts describe a real situation only: never mention a game, NPC, player, or quest.
21
22 ## Non-negotiable runtime rule
23+
24 The game is an HTTP client of externally managed model services. It must **never** start, stop, restart, kill, configure, or download a model for `llama-server`, `audiocpp_server`, or `whisper-server`.
25
26diff --git a/README.md b/README.md
27index 3373762676b99831f2d4f1b38067926a384ecc87..e3a51b5c3dcf3345e551cc0721bd339a76bd6515 100644
28--- a/README.md
29+++ b/README.md
30@@ -40,13 +40,13 @@ check pass.
31 Every endpoint is set with a flag or an environment variable. Defaults match a
32 local setup.
33
34-| Service | Flag | Env var | Default | Use |
35-| ----------------- | ------------------ | ----------------- | --------------------------------- | ----------------------------------------------- |
36-| LLM base URL | `--llm-url` | `JP_LLM_BASE_URL` | `http://127.0.0.1:8081/v1` | `POST /chat/completions` |
37-| TTS base URL | `--tts-url` | `JP_TTS_BASE_URL` | `http://127.0.0.1:8080/v1` | `POST /audio/speech` |
38-| STT inference URL | `--stt-url` | `JP_STT_URL` | `http://127.0.0.1:8178/inference` | multipart `POST` |
39-| STT language | `--stt-language` | `JP_STT_LANGUAGE` | `auto` | optional Whisper request language |
40-| Recorder command | `--record-command` | — | `arecord` | 16 kHz mono S16_LE WAV capture (see note above) |
41+| Service | Flag | Env var | Default | Use |
42+| ----------------- | ------------------ | ----------------- | ----------------------------------------- | ----------------------------------------------- |
43+| LLM base URL | `--llm-url` | `JP_LLM_BASE_URL` | `https://llama.home.theedgeofrage.com/v1` | `POST /chat/completions` |
44+| TTS base URL | `--tts-url` | `JP_TTS_BASE_URL` | `http://127.0.0.1:8080/v1` | `POST /audio/speech` |
45+| STT inference URL | `--stt-url` | `JP_STT_URL` | `http://127.0.0.1:8178/inference` | multipart `POST` |
46+| STT language | `--stt-language` | `JP_STT_LANGUAGE` | `auto` | optional Whisper request language |
47+| Recorder command | `--record-command` | — | `arecord` | 16 kHz mono S16_LE WAV capture (see note above) |
48
49 Content paths:
50
51diff --git a/docs/services.md b/docs/services.md
52index 7139cb5a0c002c8af1890170e3779a68a746c786..39a4edd3c228d5b47ca4b93693fe7b1083e3e096 100644
53--- a/docs/services.md
54+++ b/docs/services.md
55@@ -9,13 +9,13 @@ All endpoints are configuration, not hard-coded process assumptions. Each can be
56 set with a flag (see `jp run -help`) or an environment variable, and defaults
57 match the local setup:
58
59-| Service | Flag | Env var | Default | Use |
60-| --- | --- | --- | --- | --- |
61-| LLM base URL | `--llm-url` | `JP_LLM_BASE_URL` | `http://127.0.0.1:8081/v1` | `POST /chat/completions` |
62-| TTS base URL | `--tts-url` | `JP_TTS_BASE_URL` | `http://127.0.0.1:8080/v1` | `POST /audio/speech` |
63-| STT inference URL | `--stt-url` | `JP_STT_URL` | `http://127.0.0.1:8178/inference` | multipart `POST` |
64-| STT language | `--stt-language` | `JP_STT_LANGUAGE` | `auto` | optional Whisper request language |
65-| Recorder command | `--record-command` | - | `arecord` | 16kHz mono S16_LE WAV capture |
66+| Service | Flag | Env var | Default | Use |
67+| ----------------- | ------------------ | ----------------- | ----------------------------------------- | --------------------------------- |
68+| LLM base URL | `--llm-url` | `JP_LLM_BASE_URL` | `https://llama.home.theedgeofrage.com/v1` | `POST /chat/completions` |
69+| TTS base URL | `--tts-url` | `JP_TTS_BASE_URL` | `http://127.0.0.1:8080/v1` | `POST /audio/speech` |
70+| STT inference URL | `--stt-url` | `JP_STT_URL` | `http://127.0.0.1:8178/inference` | multipart `POST` |
71+| STT language | `--stt-language` | `JP_STT_LANGUAGE` | `auto` | optional Whisper request language |
72+| Recorder command | `--record-command` | - | `arecord` | 16kHz mono S16_LE WAV capture |
73
74 At startup, `jp run` makes only safe, bounded HTTP readiness checks (a few
75 seconds each, no inference requests) and prints a per-service status line with
76diff --git a/internal/config/config.go b/internal/config/config.go
77index c0027bb55b90914609c6644edb11b5ec7eed6a53..6dcbc0c8b6acd3e32c2cf8bf212b1063aec4957d 100644
78--- a/internal/config/config.go
79+++ b/internal/config/config.go
80@@ -7,7 +7,7 @@ package config
81 // Config is the shared option group. Subcommands embed it in their own options
82 // struct and parse with github.com/jessevdk/go-flags.
83 type Config struct {
84- LLMBaseURL string `long:"llm-url" env:"JP_LLM_BASE_URL" default:"http://127.0.0.1:8081/v1" description:"OpenAI-compatible LLM base URL"`
85+ LLMBaseURL string `long:"llm-url" env:"JP_LLM_BASE_URL" default:"https://llama.home.theedgeofrage.com/v1" description:"OpenAI-compatible LLM base URL"`
86 TTSBaseURL string `long:"tts-url" env:"JP_TTS_BASE_URL" default:"http://127.0.0.1:8080/v1" description:"OpenAI-compatible TTS base URL"`
87 STTURL string `long:"stt-url" env:"JP_STT_URL" default:"http://127.0.0.1:8178/inference" description:"Whisper inference URL"`
88 STTLanguage string `long:"stt-language" env:"JP_STT_LANGUAGE" default:"auto" description:"Optional Whisper request language"`
89diff --git a/internal/llm/client.go b/internal/llm/client.go
90index fc6551c9454fb7afcd17a950b329ea6782e9c5cc..0a239cebc69cad200c44eaacdc72e23d56a61803 100644
91--- a/internal/llm/client.go
92+++ b/internal/llm/client.go
93@@ -15,6 +15,8 @@ import (
94 "strings"
95 )
96
97+const llmModel = "qwen3.8-27b-q3"
98+
99 type Role string
100
101 const (
102@@ -38,7 +40,7 @@ type Options struct {
103 }
104
105 func Qwen3Options() Options {
106- return Options{Model: "jp", Temperature: 0.4, MaxTokens: 256, EnableThinking: false}
107+ return Options{Model: llmModel, Temperature: 1.0, MaxTokens: 256, EnableThinking: false}
108 }
109
110 type Client struct {