Initial updates, before testing

core/control-plane/schema.ts

@@ -53,6 +53,8 @@ const modelDescriptionSchema = z.object({
       presencePenalty: z.number().optional(),
       frequencyPenalty: z.number().optional(),
       mirostat: z.number().optional(),
+      mirostatEta: z.number().optional(),
+      mirostatTau: z.number().optional(),
       stop: z.array(z.string()).optional(),
       maxTokens: z.number().optional(),
       numThreads: z.number().optional(),

core/index.d.ts

@@ -860,6 +860,8 @@ interface BaseCompletionOptions {
   presencePenalty?: number;
   frequencyPenalty?: number;
   mirostat?: number;
+  mirostatEta?: number;
+  mirostatTau?: number;
   stop?: string[];
   maxTokens?: number;
   numThreads?: number;

core/llm/llms/LlamaCpp.ts

@@ -15,6 +15,8 @@ class LlamaCpp extends BaseLLM {
       presence_penalty: options.presencePenalty,
       min_p: options.minP,
       mirostat: options.mirostat,
+      mirostatEta: options.mirostatEta,
+      mirostatTau: options.mirostatTau,
       stop: options.stop,
       top_k: options.topK,
       top_p: options.topP,

core/llm/llms/Ollama.ts

@@ -182,6 +182,8 @@ class Ollama extends BaseLLM {
       stop: options.stop,
       num_ctx: this.contextLength,
       mirostat: options.mirostat,
+      mirostat_eta: options.mirostatEta,
+      mirostat_tau: options.mirostatTau,
       num_thread: options.numThreads,
       use_mmap: options.useMmap,
       min_p: options.minP,

docs/docs/reference.md

@@ -152,6 +152,8 @@ Parameters that control the behavior of text generation and completion settings.
 - `presencePenalty`: Discourages the model from generating tokens that have already appeared in the output.
 - `frequencePenalty`: Penalizes tokens based on their frequency in the text, reducing repetition.
 - `mirostat`: Enables Mirostat sampling, which controls the perplexity during text generation. Supported by Ollama, LM Studio, and llama.cpp providers (default: `0`, where `0` = disabled, `1` = Mirostat, and `2` = Mirostat 2.0).
+- `mirostatEta`: Influences how quickly the algorithm responds to feedback from the generated text. A lower learning rate will result in slower adjustments, while a higher learning rate will make the algorithm more responsive. (Default: 0.1)
+- `mirostatTau`: Controls the balance between coherence and diversity of the output. A lower value will result in more focused and coherent text. (Default: 5.0)
 - `stop`: An array of stop tokens that, when encountered, will terminate the completion. Allows specifying multiple end conditions.
 - `maxTokens`: The maximum number of tokens to generate in a completion (default: `2048`).
 - `numThreads`: The number of threads used during the generation process. Available only for Ollama as `num_thread`.

docs/i18n/zh-CN/docusaurus-plugin-content-docs/current/customize/deep-dives/prompt-files.md

@@ -59,6 +59,8 @@ Write unit tests for the above selected code, following each of these instructio
 - `presencePenalty`
 - `frequencyPenalty`
 - `mirostat`
+- `mirostatEta`
+- `mirostatTau`
 - `stop`
 - `maxTokens`
 - `description`

docs/i18n/zh-CN/docusaurus-plugin-content-docs/current/reference.md

@@ -152,6 +152,8 @@ Parameters that control the behavior of text generation and completion settings.
 - `presencePenalty`: Discourages the model from generating tokens that have already appeared in the output.
 - `frequencePenalty`: Penalizes tokens based on their frequency in the text, reducing repetition.
 - `mirostat`: Enables Mirostat sampling, which controls the perplexity during text generation. Supported by Ollama, LM Studio, and llama.cpp providers (default: `0`, where `0` = disabled, `1` = Mirostat, and `2` = Mirostat 2.0).
+- `mirostatEta`: Influences how quickly the algorithm responds to feedback from the generated text. A lower learning rate will result in slower adjustments, while a higher learning rate will make the algorithm more responsive. (Default: 0.1)
+- `mirostatTau`: Controls the balance between coherence and diversity of the output. A lower value will result in more focused and coherent text. (Default: 5.0)
 - `stop`: An array of stop tokens that, when encountered, will terminate the completion. Allows specifying multiple end conditions.
 - `maxTokens`: The maximum number of tokens to generate in a completion (default: `2048`).
 - `numThreads`: The number of threads used during the generation process. Available only for Ollama as `num_thread`.

extensions/vscode/config_schema.json

@@ -43,6 +43,16 @@
           "description": "Enable Mirostat sampling, controlling perplexity during text generation (default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0). Only available for Ollama, LM Studio, and llama.cpp providers",
           "type": "number"
         },
+        "mirostatEta": {
+          "title": "Mirostat Eta",
+          "description": "The Mirostat eta value, which controls how quickly the model converges to its optimal perplexity. A higher value means faster convergence but may result in less accurate results. Only enabled when `mirostat` is also enabled. (default: 0.1)",
+          "type": "number"
+        },
+        "mirostatTau": {
+          "title": "Mirostat Tau",
+          "description": "The Mirostat tau value, which controls how quickly the model converges to its optimal perplexity. A higher value means faster convergence but may result in less accurate results. Only enabled when `mirostat` is also enabled. (default: 5.0)",
+          "type": "number"
+        },
         "stop": {
           "title": "Stop",
           "description": "The stop tokens of the completion.",

packages/config-types/src/index.ts

@@ -8,6 +8,8 @@ export const completionOptionsSchema = z.object({
   presencePenalty: z.number().optional(),
   frequencyPenalty: z.number().optional(),
   mirostat: z.number().optional(),
+  mirostatEta: z.number().optional(),
+  mirostatTau: z.number().optional(),
   stop: z.array(z.string()).optional(),
   maxTokens: z.number().optional(),
   numThreads: z.number().optional(),