[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_158":159,"_160":161},"routes/_app.x.$handle.$name",{"_98":97,"_32":45,"_106":88,"_27":107,"_13":14,"_108":-5,"_11":109,"_110":111,"_112":113,"_114":115,"_116":117,"_20":-5,"_118":-5,"_119":-5,"_120":25,"_121":122,"_123":124,"_125":-5,"_126":-5,"_127":128},"routes/_app.x.$handle.$name.branch.$branch.version.$version.diff",{"_11":12,"_13":14,"_15":16,"_17":-5,"_18":-5},"project",{"_43":44,"_32":45,"_38":46,"_47":-5,"_48":25,"_49":50,"_51":50,"_52":53,"_54":-5,"_18":55,"_56":-5,"_57":25,"_58":-5,"_59":-5,"_60":61,"_62":63,"_64":65,"_66":67,"_68":69,"_70":-5,"_71":72,"_73":72,"_74":75,"_76":75,"_77":75,"_78":72,"_79":72,"_80":81,"_82":83,"_84":85,"_86":25},"version",7,"changes",[19],"aiChatInfo","userId",{"_20":21,"_22":23,"_24":25,"_26":25,"_27":28,"_29":28},"file",{"_30":31,"_32":28,"_33":34,"_35":37,"_38":39,"_40":14,"_41":42,"_24":25},"previous",{"_30":31,"_32":28,"_33":34,"_35":36},"deleted",false,"deletedHere","path","main.ts","latestPath","fileId","b9a5b966-5a06-11f1-9151-ee650bb23af1","name","fileType","http","content","/**\n * NovelAI -> OpenAI compatible proxy\n *\n * Env vars (set them in Val Town -> Settings -> Environment Variables):\n * NAI_TOKEN : NovelAI Persistent API token (required)\n * PROXY_KEY : optional shared secret to protect this endpoint\n *\n * Endpoints:\n * GET /v1/models -> list models (no auth needed, so client dropdowns work)\n * POST /v1/chat/completions -> OAI chat (streaming & non-streaming)\n * POST /v1/completions -> OAI text completion (NAI is natively a completion model)\n *\n * /v1 prefix is optional; /models, /chat/completions, /completions also work.\n */\n\n// @ts-ignore - npm shim\nimport llama3Tokenizer from \"npm:llama3-tokenizer-js@1.2.0\";\n\nconst NAI_TEXT_API = \"https://text.novelai.net\";\nconst NAI_LEGACY_API = \"https://api.novelai.net\";\n\ntype ModelInfo = {\n nai: string;\n tokenizer: \"llama3\" | \"nerdstash\";\n bytesPerToken: 2 | 4;\n baseUrl: string;\n};\n\nconst MODEL_MAP: Record = {\n erato: { nai: \"llama-3-erato-v1\", tokenizer: \"llama3\", bytesPerToken: 4, baseUrl: NAI_TEXT_API },\n kayra: { nai: \"kayra-v1\", tokenizer: \"nerdstash\", bytesPerToken: 2, baseUrl: NAI_TEXT_API },\n clio: { nai: \"clio-v1\", tokenizer: \"nerdstash\", bytesPerToken: 2, baseUrl: NAI_LEGACY_API },\n \"llama-3-erato-v1\": { nai: \"llama-3-erato-v1\", tokenizer: \"llama3\", bytesPerToken: 4, baseUrl: NAI_TEXT_API },\n \"kayra-v1\": { nai: \"kayra-v1\", tokenizer: \"nerdstash\", bytesPerToken: 2, baseUrl: NAI_TEXT_API },\n \"clio-v1\": { nai: \"clio-v1\", tokenizer: \"nerdstash\", bytesPerToken: 2, baseUrl: NAI_LEGACY_API },\n};\n\nconst MODELS_LIST = [\n { id: \"erato\", object: \"model\", created: 1727740800, owned_by: \"novelai\" },\n { id: \"kayra\", object: \"model\", created: 1692835200, owned_by: \"novelai\" },\n { id: \"clio\", object: \"model\", created: 1684800000, owned_by: \"novelai\" },\n];\n\nconst CORS = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Headers\": \"Authorization, Content-Type, x-api-key\",\n \"Access-Control-Allow-Methods\": \"GET, POST, OPTIONS\",\n};\n\n// ---------- helpers ----------\n\nfunction tokensToB64(tokens: number[], bytesPerToken: number): string {\n const buf = new Uint8Array(tokens.length * bytesPerToken);\n const view = new DataView(buf.buffer);\n for (let i = 0; i < tokens.length; i++) {\n if (bytesPerToken === 4) view.setUint32(i * 4, tokens[i], true);\n else view.setUint16(i * 2, tokens[i], true);\n }\n let s = \"\";\n for (let i = 0; i < buf.length; i++) s += String.fromCharCode(buf[i]);\n return btoa(s);\n}\n\nfunction b64ToTokens(b64: string, bytesPerToken: number): number[] {\n const bin = atob(b64);\n const buf = new Uint8Array(bin.length);\n for (let i = 0; i < bin.length; i++) buf[i] = bin.charCodeAt(i);\n const view = new DataView(buf.buffer);\n const tokens: number[] = [];\n for (let i = 0; i < buf.length; i += bytesPerToken) {\n tokens.push(bytesPerToken === 4 ? view.getUint32(i, true) : view.getUint16(i, true));\n }\n return tokens;\n}\n\nfunction encodeForModel(prompt: string, info: ModelInfo): number[] {\n if (info.tokenizer === \"llama3\") return llama3Tokenizer.encode(prompt);\n throw new Error(`This proxy currently only supports the llama3 tokenizer (model 'erato'). Requested model needs the NerdStash tokenizer.`);\n}\n\nfunction decodeForModel(tokens: number[], info: ModelInfo): string {\n if (info.tokenizer === \"llama3\") return llama3Tokenizer.decode(tokens);\n return \"\";\n}\n\nfunction messagesToPrompt(messages: Array<{ role: string; content: string }>): { prompt: string; userName: string; botName: string } {\n // NAI's official chat format:\n // memory_information\n // ***\n // [ Style: chat ]\n //\n // User: hi\n // Assistant: hello!\n // User: ...\n // Assistant:\n // and we stop at \"\\nUser:\" so the model only emits one assistant turn.\n const userName = \"User\";\n const botName = \"Assistant\";\n\n const systemBits: string[] = [];\n const turnBits: string[] = [];\n for (const msg of messages) {\n const c = (msg.content ?? \"\").toString().trim();\n if (!c) continue;\n if (msg.role === \"system\") systemBits.push(c);\n else if (msg.role === \"user\") turnBits.push(`${userName}: ${c}`);\n else if (msg.role === \"assistant\") turnBits.push(`${botName}: ${c}`);\n }\n const memory = systemBits.length\n ? `${systemBits.join(\"\\n\")}\\n***\\n[ Style: chat ]\\n\\n`\n : `[ Style: chat ]\\n\\n`;\n const conversation = turnBits.join(\"\\n\");\n // end with \"Assistant:\" so the model continues as assistant\n const prompt = `${memory}${conversation}\\n${botName}:`;\n return { prompt, userName, botName };\n}\n\nfunction jsonResp(data: unknown, status = 200): Response {\n return new Response(JSON.stringify(data), {\n status,\n headers: { \"Content-Type\": \"application/json\", ...CORS },\n });\n}\n\nfunction oaiError(message: string, status = 400, type = \"invalid_request_error\"): Response {\n return jsonResp({ error: { message, type, code: null } }, status);\n}\n\n// ---------- handler ----------\n\nexport default async function handler(req: Request): Promise {\n if (req.method === \"OPTIONS\") return new Response(null, { headers: CORS });\n\n const url = new URL(req.url);\n // strip /v1 prefix if present\n const pathname = url.pathname.replace(/^\\/v1\\b/, \"\") || \"/\";\n\n // ----- GET /models : no auth, so dropdown population works -----\n if (req.method === \"GET\" && pathname === \"/models\") {\n return jsonResp({ object: \"list\", data: MODELS_LIST });\n }\n\n // Root info page\n if (req.method === \"GET\" && (pathname === \"/\" || pathname === \"\")) {\n return jsonResp({\n name: \"novelai-oai-proxy\",\n endpoints: [\"GET /v1/models\", \"POST /v1/chat/completions\", \"POST /v1/completions\"],\n models: MODELS_LIST.map((m) => m.id),\n note: \"Set NAI_TOKEN env var on the proxy. Use 'erato' for best results (only Llama3 tokenizer is bundled).\",\n });\n }\n\n // ----- everything below requires auth -----\n const proxyKey = Deno.env.get(\"PROXY_KEY\");\n if (proxyKey) {\n const auth = req.headers.get(\"Authorization\") ?? \"\";\n const tok = auth.replace(/^Bearer\\s+/i, \"\");\n if (tok !== proxyKey) return oaiError(\"Incorrect API key provided.\", 401, \"invalid_request_error\");\n }\n\n const naiToken = Deno.env.get(\"NAI_TOKEN\");\n if (!naiToken) return oaiError(\"NAI_TOKEN env var is not configured on the proxy.\", 500, \"server_error\");\n\n // ----- POST /chat/completions or /completions -----\n if (req.method === \"POST\" && (pathname === \"/chat/completions\" || pathname === \"/completions\")) {\n const isChat = pathname === \"/chat/completions\";\n let body: any;\n try { body = await req.json(); } catch { return oaiError(\"Invalid JSON body.\"); }\n\n const requestedModel = (body.model ?? \"erato\").toString();\n const info = MODEL_MAP[requestedModel];\n if (!info) return oaiError(`Unknown model '${requestedModel}'. Available: ${Object.keys(MODEL_MAP).join(\", \")}`);\n if (info.tokenizer !== \"llama3\") {\n return oaiError(\n `Model '${requestedModel}' uses NerdStash tokenizer which isn't bundled in this proxy yet. Use 'erato' instead.`,\n 400,\n );\n }\n\n const promptStr = isChat\n ? messagesToPrompt(body.messages ?? []).prompt\n : (body.prompt ?? \"\").toString();\n if (!promptStr) return oaiError(\"Empty prompt.\");\n\n let promptTokens: number[];\n try { promptTokens = encodeForModel(promptStr, info); } catch (e) { return oaiError(String(e), 500); }\n\n // stop sequence: \"\\nUser:\" so the bot stops before generating the next user turn (chat mode only)\n let stopSequences: number[][] | undefined;\n if (isChat) {\n const stopStr = \"\\nUser:\";\n const stopToks = encodeForModel(stopStr, info);\n stopSequences = [stopToks];\n }\n\n const stream = !!body.stream;\n const maxTokens = Math.min(Math.max(body.max_tokens ?? 200, 1), 2048);\n const temperature = body.temperature ?? 1.0;\n const topP = body.top_p ?? 0.95;\n\n const naiBody = {\n input: tokensToB64(promptTokens, info.bytesPerToken),\n model: info.nai,\n parameters: {\n use_string: false,\n temperature,\n max_length: maxTokens,\n min_length: 1,\n top_p: topP,\n top_k: 0,\n top_a: 1.0,\n typical_p: 1.0,\n tail_free_sampling: 1.0,\n repetition_penalty: 1.0,\n repetition_penalty_range: 2048,\n repetition_penalty_slope: 0.0,\n repetition_penalty_frequency: 0.0,\n repetition_penalty_presence: 0.0,\n generate_until_sentence: true,\n use_cache: false,\n return_full_text: false,\n prefix: \"vanilla\",\n },\n };\n\n const endpoint = `${info.baseUrl}/ai/generate${stream ? \"-stream\" : \"\"}`;\n const naiRes = await fetch(endpoint, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${naiToken}`,\n \"Content-Type\": \"application/json\",\n \"Accept\": stream ? \"text/event-stream\" : \"application/json\",\n \"User-Agent\": \"Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36\",\n \"Origin\": \"https://novelai.net\",\n \"Referer\": \"https://novelai.net/\",\n },\n body: JSON.stringify(naiBody),\n });\n\n if (!naiRes.ok) {\n const txt = await naiRes.text();\n return oaiError(`NovelAI upstream error (${naiRes.status}): ${txt}`, naiRes.status, \"upstream_error\");\n }\n\n const completionId = `chatcmpl-nai-${Date.now()}`;\n const created = Math.floor(Date.now() / 1000);\n\n if (stream) {\n return streamResponse(naiRes, requestedModel, info, completionId, created, isChat);\n }\n\n // non-stream\n const naiJson = await naiRes.json();\n const outTokens = b64ToTokens(naiJson.output ?? \"\", info.bytesPerToken);\n const text = decodeForModel(outTokens, info);\n\n if (isChat) {\n return jsonResp({\n id: completionId, object: \"chat.completion\", created, model: requestedModel,\n choices: [{ index: 0, message: { role: \"assistant\", content: text }, finish_reason: \"stop\" }],\n usage: { prompt_tokens: promptTokens.length, completion_tokens: outTokens.length, total_tokens: promptTokens.length + outTokens.length },\n });\n } else {\n return jsonResp({\n id: completionId, object: \"text_completion\", created, model: requestedModel,\n choices: [{ text, index: 0, finish_reason: \"stop\", logprobs: null }],\n usage: { prompt_tokens: promptTokens.length, completion_tokens: outTokens.length, total_tokens: promptTokens.length + outTokens.length },\n });\n }\n }\n\n return oaiError(`No such endpoint: ${req.method} ${pathname}`, 404, \"not_found\");\n}\n\n// ---------- streaming ----------\n\nfunction streamResponse(\n naiRes: Response, modelId: string, info: ModelInfo, id: string, created: number, isChat: boolean,\n): Response {\n const { readable, writable } = new TransformStream();\n const writer = writable.getWriter();\n const encoder = new TextEncoder();\n\n (async () => {\n const reader = naiRes.body!.getReader();\n const decoder = new TextDecoder();\n let buf = \"\";\n const allTokens: number[] = [];\n let lastText = \"\";\n\n try {\n // initial role chunk for chat\n if (isChat) {\n const chunk = {\n id, object: \"chat.completion.chunk\", created, model: modelId,\n choices: [{ index: 0, delta: { role: \"assistant\" }, finish_reason: null }],\n };\n await writer.write(encoder.encode(`data: ${JSON.stringify(chunk)}\\n\\n`));\n }\n\n while (true) {\n const { done, value } = await reader.read();\n if (done) break;\n buf += decoder.decode(value, { stream: true });\n const lines = buf.split(\"\\n\");\n buf = lines.pop() ?? \"\";\n for (const raw of lines) {\n const line = raw.trim();\n if (!line.startsWith(\"data:\")) continue;\n const data = line.slice(5).trim();\n if (!data || data === \"[DONE]\") continue;\n try {\n const parsed = JSON.parse(data);\n const tokB64 = parsed.token;\n if (!tokB64) continue;\n const newToks = b64ToTokens(tokB64, info.bytesPerToken);\n allTokens.push(...newToks);\n const fullText = decodeForModel(allTokens, info);\n const delta = fullText.slice(lastText.length);\n if (!delta) continue;\n lastText = fullText;\n const chunk = isChat\n ? { id, object: \"chat.completion.chunk\", created, model: modelId, choices: [{ index: 0, delta: { content: delta }, finish_reason: null }] }\n : { id, object: \"text_completion\", created, model: modelId, choices: [{ text: delta, index: 0, finish_reason: null, logprobs: null }] };\n await writer.write(encoder.encode(`data: ${JSON.stringify(chunk)}\\n\\n`));\n } catch { /* skip malformed */ }\n }\n }\n\n const finishChunk = isChat\n ? { id, object: \"chat.completion.chunk\", created, model: modelId, choices: [{ index: 0, delta: {}, finish_reason: \"stop\" }] }\n : { id, object: \"text_completion\", created, model: modelId, choices: [{ text: \"\", index: 0, finish_reason: \"stop\", logprobs: null }] };\n await writer.write(encoder.encode(`data: ${JSON.stringify(finishChunk)}\\n\\n`));\n await writer.write(encoder.encode(\"data: [DONE]\\n\\n\"));\n } catch (e) {\n try {\n await writer.write(encoder.encode(`data: ${JSON.stringify({ error: { message: String(e), type: \"stream_error\" } })}\\n\\n`));\n } catch {}\n } finally {\n try { await writer.close(); } catch {}\n }\n })();\n\n return new Response(readable, {\n headers: { \"Content-Type\": \"text/event-stream\", \"Cache-Control\": \"no-cache\", \"Connection\": \"keep-alive\", ...CORS },\n });\n}\n","/**\n * NovelAI -> OpenAI compatible proxy\n *\n * Env vars (set them in Val Town -> Settings -> Environment Variables):\n * NAI_TOKEN : NovelAI Persistent API token (required)\n * PROXY_KEY : optional shared secret to protect this endpoint\n *\n * Endpoints:\n * GET /v1/models -> list models (no auth needed, so client dropdowns work)\n * POST /v1/chat/completions -> OAI chat (streaming & non-streaming)\n * POST /v1/completions -> OAI text completion (NAI is natively a completion model)\n *\n * /v1 prefix is optional; /models, /chat/completions, /completions also work.\n */\n\n// @ts-ignore - npm shim\nimport llama3Tokenizer from \"npm:llama3-tokenizer-js@1.2.0\";\n\nconst NAI_TEXT_API = \"https://text.novelai.net\";\nconst NAI_LEGACY_API = \"https://api.novelai.net\";\n\ntype ModelInfo = {\n nai: string;\n tokenizer: \"llama3\" | \"nerdstash\";\n bytesPerToken: 2 | 4;\n baseUrl: string;\n};\n\nconst MODEL_MAP: Record = {\n erato: { nai: \"llama-3-erato-v1\", tokenizer: \"llama3\", bytesPerToken: 4, baseUrl: NAI_TEXT_API },\n kayra: { nai: \"kayra-v1\", tokenizer: \"nerdstash\", bytesPerToken: 2, baseUrl: NAI_TEXT_API },\n clio: { nai: \"clio-v1\", tokenizer: \"nerdstash\", bytesPerToken: 2, baseUrl: NAI_LEGACY_API },\n \"llama-3-erato-v1\": { nai: \"llama-3-erato-v1\", tokenizer: \"llama3\", bytesPerToken: 4, baseUrl: NAI_TEXT_API },\n \"kayra-v1\": { nai: \"kayra-v1\", tokenizer: \"nerdstash\", bytesPerToken: 2, baseUrl: NAI_TEXT_API },\n \"clio-v1\": { nai: \"clio-v1\", tokenizer: \"nerdstash\", bytesPerToken: 2, baseUrl: NAI_LEGACY_API },\n};\n\nconst MODELS_LIST = [\n { id: \"erato\", object: \"model\", created: 1727740800, owned_by: \"novelai\" },\n { id: \"kayra\", object: \"model\", created: 1692835200, owned_by: \"novelai\" },\n { id: \"clio\", object: \"model\", created: 1684800000, owned_by: \"novelai\" },\n];\n\nconst CORS = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Headers\": \"Authorization, Content-Type, x-api-key\",\n \"Access-Control-Allow-Methods\": \"GET, POST, OPTIONS\",\n};\n\n// ---------- helpers ----------\n\nfunction tokensToB64(tokens: number[], bytesPerToken: number): string {\n const buf = new Uint8Array(tokens.length * bytesPerToken);\n const view = new DataView(buf.buffer);\n for (let i = 0; i < tokens.length; i++) {\n if (bytesPerToken === 4) view.setUint32(i * 4, tokens[i], true);\n else view.setUint16(i * 2, tokens[i], true);\n }\n let s = \"\";\n for (let i = 0; i < buf.length; i++) s += String.fromCharCode(buf[i]);\n return btoa(s);\n}\n\nfunction b64ToTokens(b64: string, bytesPerToken: number): number[] {\n const bin = atob(b64);\n const buf = new Uint8Array(bin.length);\n for (let i = 0; i < bin.length; i++) buf[i] = bin.charCodeAt(i);\n const view = new DataView(buf.buffer);\n const tokens: number[] = [];\n for (let i = 0; i < buf.length; i += bytesPerToken) {\n tokens.push(bytesPerToken === 4 ? view.getUint32(i, true) : view.getUint16(i, true));\n }\n return tokens;\n}\n\nfunction encodeForModel(prompt: string, info: ModelInfo): number[] {\n if (info.tokenizer === \"llama3\") return llama3Tokenizer.encode(prompt);\n throw new Error(`This proxy currently only supports the llama3 tokenizer (model 'erato'). Requested model needs the NerdStash tokenizer.`);\n}\n\nfunction decodeForModel(tokens: number[], info: ModelInfo): string {\n if (info.tokenizer === \"llama3\") return llama3Tokenizer.decode(tokens);\n return \"\";\n}\n\nfunction messagesToPrompt(messages: Array<{ role: string; content: string }>): { prompt: string; userName: string; botName: string } {\n // NAI's official chat format:\n // memory_information\n // ***\n // [ Style: chat ]\n //\n // User: hi\n // Assistant: hello!\n // User: ...\n // Assistant:\n // and we stop at \"\\nUser:\" so the model only emits one assistant turn.\n const userName = \"User\";\n const botName = \"Assistant\";\n\n const systemBits: string[] = [];\n const turnBits: string[] = [];\n for (const msg of messages) {\n const c = (msg.content ?? \"\").toString().trim();\n if (!c) continue;\n if (msg.role === \"system\") systemBits.push(c);\n else if (msg.role === \"user\") turnBits.push(`${userName}: ${c}`);\n else if (msg.role === \"assistant\") turnBits.push(`${botName}: ${c}`);\n }\n const memory = systemBits.length\n ? `${systemBits.join(\"\\n\")}\\n***\\n[ Style: chat ]\\n\\n`\n : `[ Style: chat ]\\n\\n`;\n const conversation = turnBits.join(\"\\n\");\n // end with \"Assistant:\" so the model continues as assistant\n const prompt = `${memory}${conversation}\\n${botName}:`;\n return { prompt, userName, botName };\n}\n\nfunction jsonResp(data: unknown, status = 200): Response {\n return new Response(JSON.stringify(data), {\n status,\n headers: { \"Content-Type\": \"application/json\", ...CORS },\n });\n}\n\nfunction oaiError(message: string, status = 400, type = \"invalid_request_error\"): Response {\n return jsonResp({ error: { message, type, code: null } }, status);\n}\n\n// ---------- handler ----------\n\nexport default async function handler(req: Request): Promise {\n if (req.method === \"OPTIONS\") return new Response(null, { headers: CORS });\n\n const url = new URL(req.url);\n // strip /v1 prefix if present\n const pathname = url.pathname.replace(/^\\/v1\\b/, \"\") || \"/\";\n\n // ----- GET /models : no auth, so dropdown population works -----\n if (req.method === \"GET\" && pathname === \"/models\") {\n return jsonResp({ object: \"list\", data: MODELS_LIST });\n }\n\n // Root info page\n if (req.method === \"GET\" && (pathname === \"/\" || pathname === \"\")) {\n return jsonResp({\n name: \"novelai-oai-proxy\",\n endpoints: [\"GET /v1/models\", \"POST /v1/chat/completions\", \"POST /v1/completions\"],\n models: MODELS_LIST.map((m) => m.id),\n note: \"Set NAI_TOKEN env var on the proxy. Use 'erato' for best results (only Llama3 tokenizer is bundled).\",\n });\n }\n\n // ----- everything below requires auth -----\n const proxyKey = Deno.env.get(\"PROXY_KEY\");\n if (proxyKey) {\n const auth = req.headers.get(\"Authorization\") ?? \"\";\n const tok = auth.replace(/^Bearer\\s+/i, \"\");\n if (tok !== proxyKey) return oaiError(\"Incorrect API key provided.\", 401, \"invalid_request_error\");\n }\n\n const naiToken = Deno.env.get(\"NAI_TOKEN\");\n if (!naiToken) return oaiError(\"NAI_TOKEN env var is not configured on the proxy.\", 500, \"server_error\");\n\n // ----- POST /chat/completions or /completions -----\n if (req.method === \"POST\" && (pathname === \"/chat/completions\" || pathname === \"/completions\")) {\n const isChat = pathname === \"/chat/completions\";\n let body: any;\n try { body = await req.json(); } catch { return oaiError(\"Invalid JSON body.\"); }\n\n const requestedModel = (body.model ?? \"erato\").toString();\n const info = MODEL_MAP[requestedModel];\n if (!info) return oaiError(`Unknown model '${requestedModel}'. Available: ${Object.keys(MODEL_MAP).join(\", \")}`);\n if (info.tokenizer !== \"llama3\") {\n return oaiError(\n `Model '${requestedModel}' uses NerdStash tokenizer which isn't bundled in this proxy yet. Use 'erato' instead.`,\n 400,\n );\n }\n\n const promptStr = isChat\n ? messagesToPrompt(body.messages ?? []).prompt\n : (body.prompt ?? \"\").toString();\n if (!promptStr) return oaiError(\"Empty prompt.\");\n\n let promptTokens: number[];\n try { promptTokens = encodeForModel(promptStr, info); } catch (e) { return oaiError(String(e), 500); }\n\n // stop sequence: \"\\nUser:\" so the bot stops before generating the next user turn (chat mode only)\n let stopSequences: number[][] | undefined;\n if (isChat) {\n const stopStr = \"\\nUser:\";\n const stopToks = encodeForModel(stopStr, info);\n stopSequences = [stopToks];\n }\n\n const stream = !!body.stream;\n const maxTokens = Math.min(Math.max(body.max_tokens ?? 200, 1), 2048);\n const temperature = body.temperature ?? 1.0;\n const topP = body.top_p ?? 0.95;\n\n const naiBody: any = {\n input: tokensToB64(promptTokens, info.bytesPerToken),\n model: info.nai,\n parameters: {\n use_string: false,\n temperature,\n max_length: maxTokens,\n min_length: 1,\n top_p: topP,\n top_k: 0,\n top_a: 1.0,\n typical_p: 1.0,\n tail_free_sampling: 1.0,\n repetition_penalty: 1.0,\n repetition_penalty_range: 2048,\n repetition_penalty_slope: 0.0,\n repetition_penalty_frequency: 0.0,\n repetition_penalty_presence: 0.0,\n generate_until_sentence: true,\n use_cache: false,\n return_full_text: false,\n prefix: \"vanilla\",\n },\n };\n if (stopSequences) naiBody.parameters.stop_sequences = stopSequences;\n\n const endpoint = `${info.baseUrl}/ai/generate${stream ? \"-stream\" : \"\"}`;\n const naiRes = await fetch(endpoint, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${naiToken}`,\n \"Content-Type\": \"application/json\",\n \"Accept\": stream ? \"text/event-stream\" : \"application/json\",\n \"User-Agent\": \"Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36\",\n \"Origin\": \"https://novelai.net\",\n \"Referer\": \"https://novelai.net/\",\n },\n body: JSON.stringify(naiBody),\n });\n\n if (!naiRes.ok) {\n const txt = await naiRes.text();\n return oaiError(`NovelAI upstream error (${naiRes.status}): ${txt}`, naiRes.status, \"upstream_error\");\n }\n\n const completionId = `chatcmpl-nai-${Date.now()}`;\n const created = Math.floor(Date.now() / 1000);\n\n if (stream) {\n return streamResponse(naiRes, requestedModel, info, completionId, created, isChat);\n }\n\n // non-stream\n const naiJson = await naiRes.json();\n const outTokens = b64ToTokens(naiJson.output ?? \"\", info.bytesPerToken);\n const text = decodeForModel(outTokens, info);\n\n if (isChat) {\n return jsonResp({\n id: completionId, object: \"chat.completion\", created, model: requestedModel,\n choices: [{ index: 0, message: { role: \"assistant\", content: text }, finish_reason: \"stop\" }],\n usage: { prompt_tokens: promptTokens.length, completion_tokens: outTokens.length, total_tokens: promptTokens.length + outTokens.length },\n });\n } else {\n return jsonResp({\n id: completionId, object: \"text_completion\", created, model: requestedModel,\n choices: [{ text, index: 0, finish_reason: \"stop\", logprobs: null }],\n usage: { prompt_tokens: promptTokens.length, completion_tokens: outTokens.length, total_tokens: promptTokens.length + outTokens.length },\n });\n }\n }\n\n return oaiError(`No such endpoint: ${req.method} ${pathname}`, 404, \"not_found\");\n}\n\n// ---------- streaming ----------\n\nfunction streamResponse(\n naiRes: Response, modelId: string, info: ModelInfo, id: string, created: number, isChat: boolean,\n): Response {\n const { readable, writable } = new TransformStream();\n const writer = writable.getWriter();\n const encoder = new TextEncoder();\n\n (async () => {\n const reader = naiRes.body!.getReader();\n const decoder = new TextDecoder();\n let buf = \"\";\n const allTokens: number[] = [];\n let lastText = \"\";\n\n try {\n // initial role chunk for chat\n if (isChat) {\n const chunk = {\n id, object: \"chat.completion.chunk\", created, model: modelId,\n choices: [{ index: 0, delta: { role: \"assistant\" }, finish_reason: null }],\n };\n await writer.write(encoder.encode(`data: ${JSON.stringify(chunk)}\\n\\n`));\n }\n\n while (true) {\n const { done, value } = await reader.read();\n if (done) break;\n buf += decoder.decode(value, { stream: true });\n const lines = buf.split(\"\\n\");\n buf = lines.pop() ?? \"\";\n for (const raw of lines) {\n const line = raw.trim();\n if (!line.startsWith(\"data:\")) continue;\n const data = line.slice(5).trim();\n if (!data || data === \"[DONE]\") continue;\n try {\n const parsed = JSON.parse(data);\n const tokB64 = parsed.token;\n if (!tokB64) continue;\n const newToks = b64ToTokens(tokB64, info.bytesPerToken);\n allTokens.push(...newToks);\n const fullText = decodeForModel(allTokens, info);\n const delta = fullText.slice(lastText.length);\n if (!delta) continue;\n lastText = fullText;\n const chunk = isChat\n ? { id, object: \"chat.completion.chunk\", created, model: modelId, choices: [{ index: 0, delta: { content: delta }, finish_reason: null }] }\n : { id, object: \"text_completion\", created, model: modelId, choices: [{ text: delta, index: 0, finish_reason: null, logprobs: null }] };\n await writer.write(encoder.encode(`data: ${JSON.stringify(chunk)}\\n\\n`));\n } catch { /* skip malformed */ }\n }\n }\n\n const finishChunk = isChat\n ? { id, object: \"chat.completion.chunk\", created, model: modelId, choices: [{ index: 0, delta: {}, finish_reason: \"stop\" }] }\n : { id, object: \"text_completion\", created, model: modelId, choices: [{ text: \"\", index: 0, finish_reason: \"stop\", logprobs: null }] };\n await writer.write(encoder.encode(`data: ${JSON.stringify(finishChunk)}\\n\\n`));\n await writer.write(encoder.encode(\"data: [DONE]\\n\\n\"));\n } catch (e) {\n try {\n await writer.write(encoder.encode(`data: ${JSON.stringify({ error: { message: String(e), type: \"stream_error\" } })}\\n\\n`));\n } catch {}\n } finally {\n try { await writer.close(); } catch {}\n }\n })();\n\n return new Response(readable, {\n headers: { \"Content-Type\": \"text/event-stream\", \"Cache-Control\": \"no-cache\", \"Connection\": \"keep-alive\", ...CORS },\n });\n}\n","createdAt",["D",1779912790632],"branchVersion","branchId","b9a1d10c-5a06-11f1-9151-ee650bb23af1","id","b9a12162-5a06-11f1-9151-ee650bb23af1","novelai-oai-proxy",["D",1779912035584],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","NovelAI → OpenAI compatible proxy","imageUrl","a2219a2d-5524-4b53-9a3b-541c4598b16f","pinnedFileId","pinPreview","pinMigrated","screenshotVersion","likeCount",0,"orgId","019e4502-4947-77cb-974b-8d134448155b","tursoName","b9a121625a0611f19151ee650bb23af1","tursoHostname","b9a121625a0611f19151ee650bb23af1-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789510872811],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_98":97,"_105":101},"org",{"_98":97,"_99":97,"_100":101,"_18":55,"_102":103,"_104":25},"branch",{"_13":87,"_32":88,"_43":42,"_89":44,"_90":-5,"_91":-5,"_92":-5,"_93":-5,"_94":-5,"_95":45,"_96":97},"isOwned",22,"main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","swdu","handle","displayName","avatar","https://imagedelivery.net/iHX6Ovru0O7AjmyT5yZRoA/a09a64c9-bba7-4c9a-b0b8-81a1df1e0a00/square512?exp=1789527599&sig=fc4c31b0c9c05164cd546285819bd5bde947be41b78fcb513b66e1828078f811","tier","free","canRestrictVals","profileImageUrl","branchName","","adoptOrgId",{"_43":44,"_32":45,"_38":153,"_47":-5,"_48":25,"_49":50,"_51":50,"_52":53,"_54":-5,"_18":55,"_56":-5,"_57":25,"_58":-5,"_59":-5,"_60":61,"_62":63,"_64":65,"_66":67,"_68":154,"_70":-5,"_71":72,"_73":72,"_74":75,"_76":75,"_77":75,"_78":72,"_79":72,"_80":155,"_82":156,"_84":157,"_86":25},"tree",[148],"files",[148],"counts",{"_137":61,"_138":139,"_140":61,"_141":142,"_143":144},"contents",[129],"upstreamProject","upstreamBranch","emptyEnvVars","flags",[],"hasHttpFiles",true,"grants","orgs","customDomains",{},{"_43":130,"_32":28,"_30":31,"_131":-5,"_41":42,"_40":14,"_132":25,"_38":133,"_13":134,"_33":34,"_35":37,"_24":25,"_135":-5,"_27":136},"7bad932a-5a08-11f1-a13e-ee650bb23af1","parentId","latest",["D",1779912790632],6,"label",[28],"pulls","branches",1,"forks","envVars",{"_145":139,"_147":61},"connections",{"_145":61,"_146":61},"count","unfilledCount","emptyCount",{"_43":130,"_32":28,"_30":31,"_131":-5,"_41":42,"_40":14,"_132":25,"_38":149,"_13":134,"_33":34,"_35":37,"_24":25,"_135":-5,"_150":151,"_27":152},["D",1779912790632],"children",[],[28],["D",1779912035584],["D",1789510872811],{"_98":97,"_105":101},{"_98":97,"_99":97,"_100":101,"_18":55,"_102":103,"_104":25},{"_13":87,"_32":88,"_43":42,"_89":44,"_90":-5,"_91":-5,"_92":-5,"_93":-5,"_94":-5,"_95":45,"_96":97},"ctx",{"_80":-5,"_126":191,"_192":193,"_194":195,"_196":25,"_197":25,"_198":25,"_199":124,"_200":-5,"_201":-7,"_202":-7,"_203":61,"_204":25,"_205":25,"_206":-5,"_207":25,"_208":25,"_209":25,"_210":-5},"ENV",{"_162":163,"_164":165,"_166":88,"_167":168,"_169":170,"_171":172,"_173":174,"_175":176,"_177":178,"_179":180,"_181":182,"_183":184,"_185":186,"_187":188,"_189":190},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","e87eb04e707cf64e7e532c8e8ec395f209e17f15","RENDER_GIT_BRANCH","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]

Diff · v7