mirror of
https://github.com/karpathy/nanochat.git
synced 2026-05-09 17:30:14 +00:00
1d2a76eec4
* feat(inference): deploy d24 SFT weights to Modal Repoint Modal inference app from the broken d20 checkpoint to our own ManmohanSharma/nanochat-d24 SFT step 484. Rewrites the standalone model as an inference-only port of nanochat/gpt.py so the modern architecture (smear gate, per-layer value embeddings, ve_gate, backout, sliding window attention via SDPA, rotary base 100000, padded vocab, logit softcap) loads cleanly from the checkpoint. Tokenizer loads the pickled tiktoken encoding directly so special tokens end up at their true IDs (32759-32767), and the stop check uses that set instead of hardcoded 0-8. GPU bumped to L4 for headroom. HF token sourced from the 'huggingface' Modal secret. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(frontend): polished redesign with serif display + dark mode Lifts the craft level of the landing and chat UI without changing the desi identity. Adds Fraunces for display headlines, a floating pill LandingNav, a saffron-glow hero with a large serif headline and black pill CTAs, and three gradient-tiled feature cards with inline SVG glyphs replacing the emoji cards. The chat empty state is now a serif greeting with pill-chip prompt starters, and ChatInput is a single rounded pod so the send button sits inside the input (fixes the misaligned floating button). Adds a class-based dark mode across the chat surfaces with a sun/moon toggle in the sidebar footer, powered by a small useTheme hook and a no-flash init script in the root layout. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore(frontend): add ESLint config so CI lint step passes next lint was failing with an interactive prompt because the repo had no ESLint config. Adds a minimal next/core-web-vitals extends and drops the now-unloadable @typescript-eslint/no-explicit-any disable directive in the stream proxy by narrowing the body type to unknown. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|---|---|---|
| .. | ||
| app | ||
| components | ||
| hooks | ||
| lib | ||
| public | ||
| store | ||
| types | ||
| .dockerignore | ||
| .env.example | ||
| .eslintrc.json | ||
| .gitignore | ||
| auth.ts | ||
| Dockerfile | ||
| index.html | ||
| middleware.ts | ||
| next-env.d.ts | ||
| next.config.mjs | ||
| package-lock.json | ||
| package.json | ||
| postcss.config.mjs | ||
| README.md | ||
| tailwind.config.ts | ||
| tsconfig.json | ||
samosaChaat — Frontend Service
Next.js 14 (App Router) application that replaces the legacy nanochat/ui.html. It blends Sarvam.ai's clean split-screen layout with samosaChaat's warm desi personality — samosa and chai illustrations, lemon-mirchi toran, gold + cream palette.
Pages
/Landing — Devanagari + Great Vibes calligraphy, animated samosa (float), chai kettle (wobble + steam), toran (pendulum), ambient doodles, CTA./loginSign-in — left: architectural/mandala motif with saffron-cream gradient; right: Google + GitHub OAuth, disabled email input./chatChat — collapsible 260px sidebar (grouped history, model selector, user avatar, logout), main area with empty state + suggestion chips, markdown rendering with code-copy, auto-expanding textarea, slash commands (/temperature,/topk,/clear,/help), SSE streaming with steam typing indicator.
Tech
- Next.js 14 (App Router, standalone output)
- Tailwind CSS (theme tokens carried from
ui.html) - Zustand (persisted conversations/settings)
- NextAuth.js v5 (Google + GitHub)
- Framer Motion (hero transitions)
- Lucide React (icons)
- react-markdown + rehype-highlight (assistant rendering)
Environment
Copy .env.example → .env.local and fill in:
| Var | Purpose |
|---|---|
NEXT_PUBLIC_APP_URL |
Public URL (defaults to http://localhost:3000) |
AUTH_SERVICE_URL |
Auth microservice (BFF only; reserved for future) |
CHAT_API_URL |
Upstream chat service. If unset the frontend serves mock echo responses — perfect for local dev. |
NEXTAUTH_SECRET |
openssl rand -base64 32 |
GOOGLE_CLIENT_ID / GOOGLE_CLIENT_SECRET |
Google OAuth (optional in dev) |
GITHUB_CLIENT_ID / GITHUB_CLIENT_SECRET |
GitHub OAuth (optional in dev) |
Middleware protects /chat/* — unauthenticated users are redirected to /login?callbackUrl=/chat.
Run locally
cd services/frontend
npm install
cp .env.example .env.local # then edit
npm run dev
# → http://localhost:3000
Without any OAuth keys the /login buttons will show the NextAuth error page; the rest of the UI works via the mocked API routes.
API routes (BFF pattern)
All client calls hit Next.js routes — the frontend never talks to the chat backend directly.
GET /api/health— service info + upstream config flagsGET /api/conversations— mocked conversation listPOST /api/chat/stream— proxies toCHAT_API_URL/chat/completionsif set, otherwise streams a mock echo. Emits SSE:data: {"token": "...", "gpu": 0}\n\nand terminates withdata: {"done": true}\n\n.
Docker
docker build -t samosachaat-frontend .
docker run --rm -p 3000:3000 \
-e NEXTAUTH_SECRET=... \
-e CHAT_API_URL=http://host.docker.internal:8000 \
samosachaat-frontend
The image is based on node:20-alpine, uses Next.js output: standalone, and runs as a non-root user.
Porting notes
All SVG assets (samosa, chai kettle, toran, steam, doodles, logo) are componentized under components/svg/. CSS keyframes (pendulum, float, wobble, steamFloat, steamType) moved into tailwind.config.ts — reference via the animate-* utilities. The original single-file UI lives at nanochat/ui.html for reference.