@nanogpt/private-mode 0.2.5 → 0.2.6

package/README.md

@@ -54,6 +54,7 @@ GET http://127.0.0.1:8787/v1/private-mode/attestation
 
 Supported private model IDs include:
 
+- `private/deepseek-v4-flash`
 - `private/kimi-k3`
 - `private/glm-5-1`
 - `private/glm-5-1-thinking`

package/lib/requestTransforms.js

@@ -27,6 +27,9 @@ const KIMI_K3_AUDIO_URL_TOKENS = 4_096;
 const KIMI_K3_VIDEO_TOKENS = 65_536;
 const KIMI_K3_REASONING_EFFORT_LEVELS = new Set(['low', 'high', 'max']);
 const KIMI_K3_TEXT_ENCODER = new TextEncoder();
+const DEEPSEEK_V4_CONTEXT_WINDOW_TOKENS = 1_048_576;
+const DEEPSEEK_V4_MAX_COMPLETION_TOKENS = 1_048_576;
+const DEEPSEEK_V4_CONTEXT_SAFETY_MARGIN_TOKENS = 10;
 const PRIVATE_TINFOIL_CHAT_COMPLETION_BODY_FIELDS = new Set([
   'chat_template_kwargs',
   'frequency_penalty',
@@ -510,15 +513,18 @@ export function normalizePrivateModelReasoningControls(body, model) {
 }
 
 export function shouldSuppressPrivateModelReasoning(body, model) {
-  if (!isPrivateModeKimiK3Model(model)) return false;
-  return body.reasoningOptOut === true ||
+  const isKimiK3 = isPrivateModeKimiK3Model(model);
+  if (!isKimiK3 && model.thinkingMode !== 'deepseek-v4') return false;
+  const reasoningVisibilityOptOut = body.reasoningOptOut === true ||
     body.exposeReasoning === false ||
     body.materializeReasoning === false ||
     coercePrivateModeBooleanFlag(body.include_reasoning) === false ||
+    (isPlainObject(body.reasoning) && body.reasoning.exclude === true);
+  return reasoningVisibilityOptOut || (isKimiK3 &&
     [body.enable_thinking, body.thinking, body.reasoning_effort, body.reasoning]
       .some((value) => isKimiK3ReasoningOptOutCandidate(value) || (
         isPlainObject(value) && value.exclude === true
-      ));
+      )));
 }
 
 function resolvePrivateModeKimiK3ReasoningEffort(body) {
@@ -580,6 +586,29 @@ function applyPrivateModeKimiK3RequestParams(body) {
   );
 }
 
+function clampPrivateModeDeepSeekV4Output(body) {
+  const requestedMaxTokens = typeof body.max_tokens === 'number' && Number.isFinite(body.max_tokens)
+    ? body.max_tokens
+    : undefined;
+  if (requestedMaxTokens === undefined) return;
+
+  const normalizedMaxTokens = requestedMaxTokens < 0
+    ? DEEPSEEK_V4_MAX_COMPLETION_TOKENS
+    : requestedMaxTokens;
+  const promptTokenEstimate = estimatePrivateModeKimiK3PromptTokens(body);
+  const remainingContext = Math.max(
+    1,
+    DEEPSEEK_V4_CONTEXT_WINDOW_TOKENS -
+      promptTokenEstimate -
+      DEEPSEEK_V4_CONTEXT_SAFETY_MARGIN_TOKENS,
+  );
+  body.max_tokens = Math.min(
+    Math.max(1, Math.floor(normalizedMaxTokens)),
+    DEEPSEEK_V4_MAX_COMPLETION_TOKENS,
+    remainingContext,
+  );
+}
+
 export function createPrivateModeReasoningContentSuppressor() {
   const openTags = ['<think>', '<thinking>', '<previous_reasoning>', '◁think▷'];
   const closeTags = ['</think>', '</thinking>', '</previous_reasoning>', '◁/think▷'];
@@ -871,6 +900,8 @@ export function applyPrivateModelRequestMutations(body, model) {
   applyTinfoilCompatibilityMutations(body, model);
   if (isPrivateModeKimiK3Model(model)) {
     applyPrivateModeKimiK3RequestParams(body);
+  } else if (model.thinkingMode === 'deepseek-v4') {
+    clampPrivateModeDeepSeekV4Output(body);
   }
 
   if (model.thinkingMode === 'gemma') {
@@ -880,16 +911,21 @@ export function applyPrivateModelRequestMutations(body, model) {
     };
     delete body.thinking;
     delete body.reasoning_effort;
-  } else if (model.thinkingMode === 'glm-5.2') {
+  } else if (
+    model.thinkingMode === 'glm-5.2' ||
+    model.thinkingMode === 'deepseek-v4'
+  ) {
     const thinkingEnabled = shouldEnableThinking(body, model);
+    const requestedReasoningEffort = body.reasoning_effort
+      ?? (isPlainObject(body.reasoning) ? body.reasoning.effort : undefined);
     body.chat_template_kwargs = {
       ...mergeChatTemplateKwargs(body),
       thinking: thinkingEnabled,
     };
 
-    if (model.thinkingMode === 'glm-5.2' && thinkingEnabled) {
+    if (thinkingEnabled) {
       body.chat_template_kwargs.reasoning_effort =
-        normalizeDeepSeekV4ReasoningEffort(body.reasoning_effort);
+        normalizeDeepSeekV4ReasoningEffort(requestedReasoningEffort);
     } else {
       delete body.chat_template_kwargs.reasoning_effort;
     }

package/models/private-tee.json

@@ -1,4 +1,16 @@
 [
+  {
+    "id": "private/deepseek-v4-flash",
+    "name": "DeepSeek V4 Flash Private",
+    "upstreamModel": "deepseek-v4-flash",
+    "billingModel": "private/deepseek-v4-flash",
+    "teeTargetModel": "deepseek-v4-flash",
+    "thinkingMode": "deepseek-v4",
+    "maxOutputTokens": 1048576,
+    "created": 1786406400,
+    "ownedBy": "nanogpt-private-mode",
+    "aliases": ["TEE/deepseek-v4-flash"]
+  },
   {
     "id": "private/kimi-k3",
     "name": "Kimi K3 Private",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nanogpt/private-mode",
-  "version": "0.2.5",
+  "version": "0.2.6",
   "description": "OpenAI-compatible localhost proxy for NanoGPT Private Mode.",
   "type": "module",
   "publishConfig": {