From 4095c68822167225a002be0b334ad408d6b128ad Mon Sep 17 00:00:00 2001
From: hobokenchicken <dustin@dustin.coffee>
Date: Tue, 7 Apr 2026 14:05:21 +0000
Subject: [PATCH] fix(ollama): improve model detection and ensure robust
 token/context limits

- Use case-insensitive matching for model names and routing
- Default max_tokens/num_predict to 8192 for all Ollama models to prevent truncation
- Increase default context window and add more large-context model families
- Ensure DeepSeek routing handles Ollama-hosted variants correctly
---
 internal/providers/ollama.go | 22 ++++++++++++++++------
 internal/server/server.go    | 21 ++++++++++++++++-----
 2 files changed, 32 insertions(+), 11 deletions(-)

diff --git a/internal/providers/ollama.go b/internal/providers/ollama.go
index 589aa4a8..2d9985e8 100644
--- a/internal/providers/ollama.go
+++ b/internal/providers/ollama.go
@@ -117,11 +117,20 @@ func BuildOllamaBody(request *models.UnifiedRequest, messagesJSON []interface{},
 	}
 
 	options := make(map[string]interface{})
+	modelLower := strings.ToLower(request.Model)
 
-	// Context window size (32k for modern models to avoid truncation)
-	if strings.Contains(request.Model, "gemma") || strings.Contains(request.Model, "llama") || strings.Contains(request.Model, "mistral") || strings.Contains(request.Model, "qwen") {
-		options["num_ctx"] = 32768
+	// Context window size (default 8k for all, 32k+ for modern large-context models)
+	ctxSize := 8192
+	if strings.Contains(modelLower, "llama3") || 
+	   strings.Contains(modelLower, "mistral") || 
+	   strings.Contains(modelLower, "mixtral") ||
+	   strings.Contains(modelLower, "qwen") || 
+	   strings.Contains(modelLower, "deepseek") ||
+	   strings.Contains(modelLower, "command-r") ||
+	   strings.Contains(modelLower, "phi3") {
+		ctxSize = 32768
 	}
+	options["num_ctx"] = ctxSize
 
 	if request.Temperature != nil {
 		body["temperature"] = *request.Temperature
@@ -131,9 +140,10 @@ func BuildOllamaBody(request *models.UnifiedRequest, messagesJSON []interface{},
 	if request.MaxTokens != nil {
 		body["max_tokens"] = *request.MaxTokens
 		options["num_predict"] = *request.MaxTokens
-	} else if strings.Contains(request.Model, "gemma") || strings.Contains(request.Model, "llama") || strings.Contains(request.Model, "mistral") || strings.Contains(request.Model, "qwen") {
-		// Default to 8192 for common Ollama models if not specified, 
-		// as Ollama's compatibility layer sometimes defaults to 128
+	} else {
+		// Default to 8192 for all Ollama models if not specified, 
+		// as Ollama's compatibility layer defaults to 128 if neither 
+		// max_tokens nor num_predict are provided.
 		body["max_tokens"] = 8192
 		options["num_predict"] = 8192
 	}
diff --git a/internal/server/server.go b/internal/server/server.go
index 4bec9944..9d429c22 100644
--- a/internal/server/server.go
+++ b/internal/server/server.go
@@ -310,15 +310,26 @@ func (s *Server) handleChatCompletions(c *gin.Context) {
 
 	// Select provider based on model name
 	providerName := "openai" // default
-	if strings.HasPrefix(req.Model, "gemini/") || strings.Contains(req.Model, "gemini") || strings.HasPrefix(req.Model, "google/") {
+	modelLower := strings.ToLower(req.Model)
+	if strings.HasPrefix(modelLower, "gemini/") || strings.Contains(modelLower, "gemini") || strings.HasPrefix(modelLower, "google/") {
 		providerName = "gemini"
-	} else if strings.HasPrefix(req.Model, "deepseek/") || strings.Contains(req.Model, "deepseek") {
+	} else if strings.HasPrefix(modelLower, "deepseek/") || (strings.Contains(modelLower, "deepseek") && !strings.Contains(modelLower, "ollama")) {
+		// Only use deepseek provider if it's not explicitly tagged for ollama
 		providerName = "deepseek"
-	} else if strings.HasPrefix(req.Model, "moonshot/") || strings.Contains(req.Model, "kimi") || strings.Contains(req.Model, "moonshot") {
+	} else if strings.HasPrefix(modelLower, "moonshot/") || strings.Contains(modelLower, "kimi") || strings.Contains(modelLower, "moonshot") {
 		providerName = "moonshot"
-	} else if strings.HasPrefix(req.Model, "grok/") || strings.Contains(req.Model, "grok") {
+	} else if strings.HasPrefix(modelLower, "grok/") || strings.Contains(modelLower, "grok") {
 		providerName = "grok"
-	} else if strings.HasPrefix(req.Model, "ollama/") || strings.Contains(req.Model, "glm-") || strings.Contains(req.Model, "qwen") || strings.Contains(req.Model, "gemma") || strings.Contains(req.Model, "llama") || strings.Contains(req.Model, "mistral") || strings.Contains(req.Model, "codellama") {
+	} else if strings.HasPrefix(modelLower, "ollama/") || 
+		strings.Contains(modelLower, "glm-") || 
+		strings.Contains(modelLower, "qwen") || 
+		strings.Contains(modelLower, "gemma") || 
+		strings.Contains(modelLower, "llama") || 
+		strings.Contains(modelLower, "mistral") || 
+		strings.Contains(modelLower, "phi") || 
+		strings.Contains(modelLower, "yi") || 
+		strings.Contains(modelLower, "codellama") || 
+		strings.Contains(modelLower, "command-r") {
 		providerName = "ollama"
 	}