Improve imatrix calibration corpus

Broaden the DS4 imatrix prompt dataset with provider-neutral agent/tool traffic, multi-language programming prompts, algorithm recall, Bash scripting, and multilingual translation tasks.

Remove duplicate rendered prompts and avoid provider-specific client references in the generated calibration corpus. This improves calibration coverage without claiming to fix a distributed GGUF bug.

gguf-tools/deepseek4-quantize.c

@@ -880,7 +880,10 @@ static expert_tensor parse_expert_tensor(const char *name) {
     expert_tensor e = {0};
     int layer = -1;
     char kind[16];
-    if (sscanf(name, "blk.%d.ffn_%15[^_]_exps.weight", &layer, kind) == 2) {
+    int rest = 0;
+    if (sscanf(name, "blk.%d.ffn_%15[^_]_exps.weight%n", &layer, kind, &rest) == 2
+        && rest == (int)strlen(name))
+    {
         if (strcmp(kind, "gate") == 0 || strcmp(kind, "down") == 0 || strcmp(kind, "up") == 0) {
             e.is_expert = true;
             e.layer = layer;

gguf-tools/imatrix/dataset/README.md

@@ -26,8 +26,11 @@ The renderer mirrors the server prompt shape:
 
 Some records include DSML tool schemas, sampled DSML tool calls, and tool-result
 turns so the imatrix sees the same special-token patterns used by agent clients.
-The corpus also includes Italian prompts, long-context source excerpts, Metal/C
-code review tasks, and inference-specific debugging tasks.
+The corpus is provider-neutral and also includes multi-language programming
+prompts, Bash scripting, algorithm recall, English-to-European-language
+translation, English-to-Chinese translation, reverse translation to English,
+long-context source excerpts, Metal/C code review tasks, and
+inference-specific debugging tasks.
 
 For normal imatrix collection, use `rendered_prompts.txt` so calibration covers
 both thinking and non-thinking modes. Split files are provided for ablations.

gguf-tools/imatrix/dataset/manifest.json

@@ -1,24 +1,30 @@
 {
-  "version": 1,
+  "version": 3,
   "purpose": "DeepSeek V4 Flash imatrix calibration prompts",
-  "record_count": 1952,
-  "rendered_utf8_bytes": 5773104,
-  "rough_token_estimate_bytes_div_4": 1443276,
+  "record_count": 3310,
+  "rendered_utf8_bytes": 10167295,
+  "rough_token_estimate_bytes_div_4": 2541823,
   "categories": {
-    "agent": 8,
+    "agent": 1106,
+    "algorithms": 32,
     "general": 40,
-    "long_context": 18,
-    "source": 1886
+    "long_context": 12,
+    "programming": 48,
+    "source": 1892,
+    "translation": 180
   },
   "modes": {
-    "nothink": 976,
-    "think": 976
+    "nothink": 1655,
+    "think": 1655
   },
   "bytes_by_category": {
-    "agent": 13100,
+    "agent": 4432105,
+    "algorithms": 9388,
     "general": 11902,
-    "long_context": 371581,
-    "source": 5376521
+    "long_context": 248628,
+    "programming": 14320,
+    "source": 5394802,
+    "translation": 56150
   },
   "files": {
     "jsonl": "prompts.jsonl",