feat: add latest qwen 2.5 coder models and adjust stop tokens

Author: carlrobertoh

src/main/java/ee/carlrobert/codegpt/completions/HuggingFaceModel.java

@@ -125,12 +125,33 @@ public enum HuggingFaceModel {
   CODE_QWEN_2_5_1_5B_Q8_0(1, 8, "Qwen2.5-Coder-1.5B-Instruct-GGUF",
       "qwen2.5-coder-1.5b-instruct-q8_0.gguf", "Qwen", 1.89),
 
+  CODE_QWEN_2_5_3B_Q4_K_M(3, 4, "Qwen2.5-Coder-3B-Instruct-GGUF",
+      "qwen2.5-coder-3b-instruct-q4_k_m.gguf", "Qwen", 2.10),
+  CODE_QWEN_2_5_3B_Q6_K(3, 6, "Qwen2.5-Coder-3B-Instruct-GGUF",
+      "qwen2.5-coder-3b-instruct-q6_k.gguf", "Qwen", 2.79),
+  CODE_QWEN_2_5_3B_Q8_0(3, 8, "Qwen2.5-Coder-3B-Instruct-GGUF",
+      "qwen2.5-coder-3b-instruct-q8_0.gguf", "Qwen", 3.62),
+
   CODE_QWEN_2_5_7B_Q4_K_M(7, 4, "Qwen2.5-Coder-7B-Instruct-GGUF",
-      "Qwen2.5-Coder-7B-Instruct-Q4_K_M.gguf", "bartowski", 4.68),
+      "qwen2.5-coder-7b-instruct-q4_k_m.gguf", "Qwen", 4.68),
   CODE_QWEN_2_5_7B_Q6_K(7, 6, "Qwen2.5-Coder-7B-Instruct-GGUF",
-      "Qwen2.5-Coder-7B-Instruct-Q6_K.gguf", "bartowski", 6.25),
+      "qwen2.5-coder-7b-instruct-q6_k.gguf", "Qwen", 6.25),
   CODE_QWEN_2_5_7B_Q8_0(7, 8, "Qwen2.5-Coder-7B-Instruct-GGUF",
-      "Qwen2.5-Coder-7B-Instruct-Q8_0.gguf", "bartowski", 8.1),
+      "qwen2.5-coder-7b-instruct-q8_0.gguf", "Qwen", 8.10),
+
+  CODE_QWEN_2_5_14B_Q4_K_M(14, 4, "Qwen2.5-Coder-14B-Instruct-GGUF",
+      "qwen2.5-coder-14b-instruct-q4_k_m.gguf", "Qwen", 8.99),
+  CODE_QWEN_2_5_14B_Q6_K(14, 6, "Qwen2.5-Coder-14B-Instruct-GGUF",
+      "qwen2.5-coder-14b-instruct-q6_k.gguf", "Qwen", 12.10),
+  CODE_QWEN_2_5_14B_Q8_0(14, 8, "Qwen2.5-Coder-14B-Instruct-GGUF",
+      "qwen2.5-coder-14b-instruct-q8_0.gguf", "Qwen", 15.70),
+
+  CODE_QWEN_2_5_32B_Q4_K_M(32, 4, "Qwen2.5-Coder-32B-Instruct-GGUF",
+      "qwen2.5-coder-32b-instruct-q4_k_m.gguf", "Qwen", 19.90),
+  CODE_QWEN_2_5_32B_Q6_K(32, 6, "Qwen2.5-Coder-32B-Instruct-GGUF",
+      "qwen2.5-coder-32b-instruct-q6_k.gguf", "Qwen", 26.90),
+  CODE_QWEN_2_5_32B_Q8_0(32, 8, "Qwen2.5-Coder-32B-Instruct-GGUF",
+      "qwen2.5-coder-32b-instruct-q8_0.gguf", "Qwen", 34.80),
 
   STABLE_CODE_3B_Q3_K_M(SC3, 3, "stable-code-instruct-3b-Q3_K_M.gguf", 1.39),
   STABLE_CODE_3B_Q4_K_M(SC3, 4, "stable-code-instruct-3b-Q4_K_M.gguf", 1.71),

src/main/java/ee/carlrobert/codegpt/completions/llama/LlamaModel.java

@@ -189,9 +189,18 @@ public enum LlamaModel {
       List.of(
           HuggingFaceModel.CODE_QWEN_2_5_1_5B_Q6_K,
           HuggingFaceModel.CODE_QWEN_2_5_1_5B_Q8_0,
+          HuggingFaceModel.CODE_QWEN_2_5_3B_Q4_K_M,
+          HuggingFaceModel.CODE_QWEN_2_5_3B_Q6_K,
+          HuggingFaceModel.CODE_QWEN_2_5_3B_Q8_0,
           HuggingFaceModel.CODE_QWEN_2_5_7B_Q4_K_M,
           HuggingFaceModel.CODE_QWEN_2_5_7B_Q6_K,
-          HuggingFaceModel.CODE_QWEN_2_5_7B_Q8_0)),
+          HuggingFaceModel.CODE_QWEN_2_5_7B_Q8_0,
+          HuggingFaceModel.CODE_QWEN_2_5_14B_Q4_K_M,
+          HuggingFaceModel.CODE_QWEN_2_5_14B_Q6_K,
+          HuggingFaceModel.CODE_QWEN_2_5_14B_Q8_0,
+          HuggingFaceModel.CODE_QWEN_2_5_32B_Q4_K_M,
+          HuggingFaceModel.CODE_QWEN_2_5_32B_Q6_K,
+          HuggingFaceModel.CODE_QWEN_2_5_32B_Q8_0)),
   STABLE_CODE(
       "Stable Code Instruct", """
       stable-code-instruct-3b is a 2.7B billion parameter decoder-only language model tuned from \

src/main/kotlin/ee/carlrobert/codegpt/codecompletions/InfillPromptTemplate.kt

@@ -50,7 +50,23 @@ enum class InfillPromptTemplate(val label: String, val stopTokens: List<String>?
             }
         }
     },
-    CODE_QWEN_2_5("CodeQwen2.5", listOf()) {
+    CODE_QWEN_2_5(
+        "CodeQwen2.5",
+        listOf(
+            "package ",
+            "import ",
+            "<|endoftext|>",
+            "<|fim_prefix|>",
+            "<|fim_middle|>",
+            "<|fim_suffix|>",
+            "<|fim_pad|>",
+            "<|cursor|>",
+            "<|repo_name|>",
+            "<|file_sep|>",
+            "<|im_start|>",
+            "<|im_end|>"
+        )
+    ) {
         override fun buildPrompt(infillDetails: InfillRequest): String {
             val infillPrompt =
                 "<|fim_prefix|> ${infillDetails.prefix} <|fim_suffix|>${infillDetails.suffix} <|fim_middle|>"
@@ -110,7 +126,8 @@ enum class InfillPromptTemplate(val label: String, val stopTokens: List<String>?
     CODESTRAL("Codestral", listOf("</s>")) {
         override fun buildPrompt(infillDetails: InfillRequest): String {
             // see https://github.com/mistralai/mistral-common/blob/master/src/mistral_common/tokens/tokenizers/base.py
-            val infillPrompt = "[SUFFIX]${infillDetails.suffix}[PREFIX]${infillDetails.prefix}[MIDDLE]"
+            val infillPrompt =
+                "[SUFFIX]${infillDetails.suffix}[PREFIX]${infillDetails.prefix}[MIDDLE]"
             return createDefaultMultiFilePrompt(infillDetails, infillPrompt)
         }
     };