Upload folder using huggingface_hub

README.md

@@ -10,7 +10,9 @@ tags:
 
 This is just, SOTA 2 and 3-bit quants for laserxtral. Not much more to it. Meow.
 
-The importance matrix, [which is generated from `20k_random_data.txt`](https://github.com/ggerganov/llama.cpp/discussions/5006#discussioncomment-8163190), is included in this repo, as `imatrix_laserxtral.dat`.
+The importance matrix, [which is generated from `group_10_merged.txt`](https://github.com/ggerganov/llama.cpp/discussions/5263#discussioncomment-8353685), is included in this repo, as `imatrix_laserxtral.dat`.
+
+***UPDATE 2/11/2024***: The models have been reuploaded, with a new importance matrix used (`group_10_merged.txt` rather than `20k_random_data.txt`), which should in theory provide better performance. I'm not an expert, don't quote me on that.
 
 ## System Prompt
 

imatrix_laserxtral.dat

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ed56f3d455e1cd26101b5e0d8df52e4fa75f0daf37c7ce4d33a9ae6af5d8809d
+oid sha256:f8c3098d4e4d585685d4d406e6674992cf86ea0142c1c3d3543714c8d08a3d1d
 size 14174650

laserxtral-IQ2_XS.gguf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6776da13949aab66f9da470ee6ab0b08c461618eb34689da941e4f66acf812c1
+oid sha256:b32b6f5d3534df494be3709568cc629fd82681b370eee441d7a213475e09736d
 size 7115310880

laserxtral-IQ2_XXS.gguf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9f8a28bde86f5d71b00d3f27eb3c34a9df96bf115e4018b5096b9593e943ea52
+oid sha256:968b84dbcb5761e7ceb8f6b7c228d4d9df914020ad1a6a35502ac46813e4c21d
 size 6402279200

laserxtral-IQ3_XXS.gguf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd8572a497d433d5e36fb0a2b7dde5203ea888ed3ed32f9fcce9f74701e72aad
+oid sha256:1ef4b65901f9c99d23df5bf9e33f0c3f6ceabed048bffd1d258b0b2113f85ed5
 size 9379079968

laserxtral-Q3_K_XS.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e2efc8d4178f78a582d3123d76e64da43033e0b75ab2a0bfa80e6189137e394
+size 9732343584

model_config.json

@@ -15,7 +15,7 @@
     "f16_kv": true,
     "use_mmap": true,
     "no_kv_offload": false,
-    "num_experts_used": 0
+    "num_experts_used": 2
   },
   "inference_params": {
     "n_threads": 4,