ollama-ai-provider - npm Package Compare versions

Comparing version 0.14.0 to 0.15.0

dist/index.d.ts

@@ -11,2 +11,18 @@ import { LanguageModelV1, ProviderV1, EmbeddingModelV1 } from '@ai-sdk/provider';
     /**
+     * Enables the use of half-precision floating point values for key-value memory. This helps in optimizing memory usage. (Default: true)
+     */
+    f16Kv?: boolean;
+    /**
+     * If set to true, reduces the VRAM usage by trading off speed for memory. (Default: false)
+     */
+    lowVram?: boolean;
+    /**
+     * Sets which GPU is the main one.
+     */
+    mainGpu?: number;
+    /**
+     * Minimum cumulative probability for tokens to be considered. (Default: 0.0)
+     */
+    minP?: number;
+    /**
      * Enable Mirostat sampling for controlling perplexity. (default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)
@@ -26,2 +42,10 @@ */
     /**
+     * Controls whether to use Non-Uniform Memory Access (NUMA) for more efficient memory management. (Default: false)
+     */
+    numa?: boolean;
+    /**
+     * Sets the number of batches to be processed. (Default: 512)
+     */
+    numBatch?: number;
+    /**
      * Sets the size of the context window used to generate the next token. (Default: 2048)
@@ -31,2 +55,22 @@ */
     /**
+     * Controls the number of GPUs to use for the operation. (Default: -1, indicates that NumGPU should be set dynamically)
+     */
+    numGpu?: number;
+    /**
+     * Keeps a number of tokens from the context. Controls how many of the previous tokens are retained. (Default: 4)
+     */
+    numKeep?: number;
+    /**
+     * Controls the number of tokens to predict in a single generation. (Default: -1)
+     */
+    numPredict?: number;
+    /**
+     * Sets the number of CPU threads to use. (Default: 0, indicates let the runtime decide)
+     */
+    numThread?: number;
+    /**
+     * Penalizes the model for generating newline characters. If set to true, it discourages the model from generating too many newlines. (Default: true)
+     */
+    penalizeNewline?: boolean;
+    /**
      * Sets how far back for the model to look back to prevent repetition. (Default: 64, 0 = disabled, -1 = num_ctx)
@@ -59,2 +103,18 @@ */
     topK?: number;
+    /**
+     * Controls the "typical" sampling probability. (Default: 1.0)
+     */
+    typicalP?: number;
+    /**
+     * Locks the memory to prevent swapping, which can be useful for performance optimization. (Default: false)
+     */
+    useMlock?: boolean;
+    /**
+     * Enables memory mapping to reduce RAM usage. (Default: false)
+     */
+    useMmap?: boolean;
+    /**
+     * If true, the model will only load the vocabulary without performing further computation. (Default: false)
+     */
+    vocabOnly?: boolean;
 }
@@ -61,0 +121,0 @@

dist/index.js

@@ -348,8 +348,18 @@ "use strict";
       options: removeUndefined({
+        f16_kv: this.settings.f16Kv,
         frequency_penalty: frequencyPenalty,
+        low_vram: this.settings.lowVram,
+        main_gpu: this.settings.mainGpu,
+        min_p: this.settings.minP,
         mirostat: this.settings.mirostat,
         mirostat_eta: this.settings.mirostatEta,
         mirostat_tau: this.settings.mirostatTau,
+        num_batch: this.settings.numBatch,
         num_ctx: this.settings.numCtx,
+        num_gpu: this.settings.numGpu,
+        num_keep: this.settings.numKeep,
         num_predict: maxTokens,
+        num_thread: this.settings.numThread,
+        numa: this.settings.numa,
+        penalize_newline: this.settings.penalizeNewline,
         presence_penalty: presencePenalty,
@@ -363,3 +373,7 @@ repeat_last_n: this.settings.repeatLastN,
         top_k: (_b = this.settings.topK) != null ? _b : topK,
-        top_p: topP
+        top_p: topP,
+        typical_p: this.settings.typicalP,
+        use_mlock: this.settings.useMlock,
+        use_mmap: this.settings.useMmap,
+        vocab_only: this.settings.vocabOnly
       })
@@ -366,0 +380,0 @@ };

package.json

 {
   "name": "ollama-ai-provider",
-  "version": "0.14.0",
+  "version": "0.15.0",
   "description": "Vercel AI Provider for running LLMs locally using Ollama",
@@ -5,0 +5,0 @@ "main": "./dist/index.js",

No alert changes

Improved metrics

Total package byte prevSize

148582

5.11%

Lines of code

1597

5.83%

No dependency changes