Added DeciLM-7B-instruct model with uniform GQA in GGUF format, in multiple quantization levels.

The follogin changes are required for the base model (7b-instruct) to work in llama.cpp codebase:

- GQA
- Uniform GQA instead of variable GQA (fixed to 4)
- ROPE
- Linear ROPE instead of dynamic ROPE scaling
- Tested on M2 with 32gb or RAM
- Reaches 12,000-13,000 tok/s

.gitattributes

@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+decilm-7b-uniform-gqa-f16.gguf filter=lfs diff=lfs merge=lfs -text
+decilm-7b-uniform-gqa-f32.gguf filter=lfs diff=lfs merge=lfs -text
+decilm-7b-uniform-gqa-q8_0.gguf filter=lfs diff=lfs merge=lfs -text

decilm-7b-uniform-gqa-f16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c28d32a79f4dea3cf931b40237b64ce9109efd7e6c8a990a8c611286b9b2341f
+size 14217232832

decilm-7b-uniform-gqa-f32.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b45f3e8a5e05b557c6f8c421d47f2a3226af912016f923150ae07789977ba4cc
+size 28430792256

decilm-7b-uniform-gqa-q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b60beae4c9d5d7131f846f7545cb386e20d8f2c613844a0142ee701537dbfe3
+size 7554187712