Rename GGUF files so the quant type ends the filename (Ollama :Q4_K_M tags)

Fixes discussion #1. Files are byte-identical; only names changed. README and SHA256SUMS.json updated.

.gitattributes

@@ -39,3 +39,6 @@ figures/calibration.png filter=lfs diff=lfs merge=lfs -text
 figures/robustness.png filter=lfs diff=lfs merge=lfs -text
 Jev-Style-v2-Q4_K_M-Calibrated.gguf filter=lfs diff=lfs merge=lfs -text
 Jev-Style-v2-BF16-Calibrated.gguf filter=lfs diff=lfs merge=lfs -text
+Jev-Style-v2-Calibrated-Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+Jev-Style-v2-Calibrated-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text
+Jev-Style-v2-Calibrated-BF16.gguf filter=lfs diff=lfs merge=lfs -text

README.md

@@ -30,9 +30,9 @@ A **Jev-style decision model** for classification, routing and typed choices. Gi
 
 | Precision | File size | Choice agreement | Macro accuracy |
 |---|---:|---:|---:|
-| [Q4_K_M](https://huggingface.co/chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF/resolve/main/Jev-Style-v2-Q4_K_M-Calibrated.gguf?download=true) | 1.27 GB | 91.4% | 78.18% |
-| [Q8_0](https://huggingface.co/chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF/resolve/main/Jev-Style-v2-Q8_0-Calibrated.gguf?download=true) | 2.01 GB | 99.2% | 78.69% |
-| [BF16](https://huggingface.co/chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF/resolve/main/Jev-Style-v2-BF16-Calibrated.gguf?download=true) | 3.78 GB | 99.6% | 79.36% |
+| [Q4_K_M](https://huggingface.co/chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF/resolve/main/Jev-Style-v2-Calibrated-Q4_K_M.gguf?download=true) | 1.27 GB | 91.4% | 78.18% |
+| [Q8_0](https://huggingface.co/chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF/resolve/main/Jev-Style-v2-Calibrated-Q8_0.gguf?download=true) | 2.01 GB | 99.2% | 78.69% |
+| [BF16](https://huggingface.co/chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF/resolve/main/Jev-Style-v2-Calibrated-BF16.gguf?download=true) | 3.78 GB | 99.6% | 79.36% |
 
 All three files include independently fitted calibration; use **runtime temperature 1.0**. Agreement is against CUDA merged BF16 on the same frozen 500-decision subset; accuracy is the task-macro average over its real-label examples. [Full precision comparison](evaluation/quantization_summary.json).
 
@@ -136,7 +136,7 @@ The benchmark figures describe the fixed CUDA reference comparison. Reliability
 python -m pip install -U huggingface_hub
 hf download chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF --local-dir jev-v2-gguf
 cd jev-v2-gguf
-llama-server -m Jev-Style-v2-Q8_0-Calibrated.gguf -c 2048 -ngl 99 --port 8080
+llama-server -m Jev-Style-v2-Calibrated-Q8_0.gguf -c 2048 -ngl 99 --port 8080
 ```
 
 In another terminal, from the same directory:
@@ -148,7 +148,7 @@ python jev_decision_client.py --url http://127.0.0.1:8080 \
   --options negative positive
 ```
 
-The quick-start command selects Q8_0. To use Q4_K_M or BF16, replace its model filename with `Jev-Style-v2-Q4_K_M-Calibrated.gguf` or `Jev-Style-v2-BF16-Calibrated.gguf`.
+The quick-start command selects Q8_0. To use Q4_K_M or BF16, replace its model filename with `Jev-Style-v2-Calibrated-Q4_K_M.gguf` or `Jev-Style-v2-Calibrated-BF16.gguf`.
 
 Use a llama.cpp build with Qwen3.5 support. Conversion and native evaluation used commit `b29c606e28a01b1bc8c1351026a0fa6e616bf6c4`. The client uses the native `/completion` endpoint, requests complete declared-option log-probabilities and increases the candidate count as needed. The supplied `gguf_logits.cpp` reads all declared-option logits directly through the C API.
 

SHA256SUMS.json

@@ -1,9 +1,9 @@
 {
-  "Jev-Style-v2-Q8_0-Calibrated.calibration.json": {
+  "Jev-Style-v2-Calibrated-Q8_0.calibration.json": {
     "bytes": 1807,
     "sha256": "7f3ae96ebf0271483365b8941124f3be9521e9e3ed4c3d62240e51c9b416ebb4"
   },
-  "Jev-Style-v2-Q8_0-Calibrated.gguf": {
+  "Jev-Style-v2-Calibrated-Q8_0.gguf": {
     "bytes": 2012004256,
     "sha256": "5c2aa0d35b24a27f03228b2c62ebaaebd9b5b785844634d4217278d822751494"
   },
@@ -13,7 +13,7 @@
   },
   "README.md": {
     "bytes": 12086,
-    "sha256": "628af2dab33d7558e3f1cca7105f9e7252a2f62b2ba9d3744b3aca3181b45604"
+    "sha256": "5d8614413dc1583b450b81ebeee569907a023c4c65d5752dd318f542b9071577"
   },
   "evaluation/baseline_sensitivity.json": {
     "bytes": 2015,
@@ -91,11 +91,11 @@
     "bytes": 147,
     "sha256": "1f819d59b9ebf6ffb79953832c3f7ae4564d7bae7fa68209b6583a9d92db2bf5"
   },
-  "Jev-Style-v2-Q4_K_M-Calibrated.gguf": {
+  "Jev-Style-v2-Calibrated-Q4_K_M.gguf": {
     "bytes": 1274388384,
     "sha256": "c697d3b29d07fdd37b6ebeb5c98066f4c31632162adca6258db23f75184eb0c4"
   },
-  "Jev-Style-v2-Q4_K_M-Calibrated.calibration.json": {
+  "Jev-Style-v2-Calibrated-Q4_K_M.calibration.json": {
     "bytes": 1824,
     "sha256": "4a5abb23d86ddd44b8b6b1054b48a0a9c40560cc0ee3edd24ae6be8e623ad5f1"
   },
@@ -107,11 +107,11 @@
     "bytes": 694,
     "sha256": "5a0e2b0696e3aa918aff6e8ad2ebf1532091edf81a10815bf3add221c36932d3"
   },
-  "Jev-Style-v2-BF16-Calibrated.gguf": {
+  "Jev-Style-v2-Calibrated-BF16.gguf": {
     "bytes": 3775700896,
     "sha256": "8baa111eec6e30a5c9e97d559b53127a19ef30171e8aed26673153b4f77adcbb"
   },
-  "Jev-Style-v2-BF16-Calibrated.calibration.json": {
+  "Jev-Style-v2-Calibrated-BF16.calibration.json": {
     "bytes": 1840,
     "sha256": "e5eb54d405d213c751ba61b334a8370d03332c915d9c49725da3b7a6ee295109"
   },