Upload folder using huggingface_hub

README.md

@@ -7,11 +7,11 @@ tags:
 - sft
 ---
 
-# selection-coder (SFT)
+# selection-coder
 
-~210M decoder, only SelectionMixing, no attention. Pretrained on code, then SFT on instruction->script.
+~210M decoder, only SelectionMixing (no attention), Muon, seq=1024. SFT on nvidia/OpenCodeInstruct.
 
-Prompt format:
+Prompt:
 ```
 ### Instruction:
 {instr}

config.json

@@ -3,11 +3,11 @@
   "d": 1024,
   "layers": 12,
   "heads": 16,
-  "seq": 512,
+  "seq": 1024,
   "ff": 4,
   "arch": "SelectionCoder",
   "mixer": "SelectionMixing(no-attention)",
-  "stage": "sft-instruct",
+  "stage": "sft-big",
   "prompt_template": "### Instruction:\n{instr}\n\n### Response:\n",
-  "step": 1161
+  "step": 1000
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dba7629d5370c770d5643ef1bbb5c2d0d605d5e8af15080c62d5b2c9b16c8783
-size 488715288
+oid sha256:cbf78e88f383aa85e463a07e66f2f612fad52bee95419c12f8e44cf22c81231a
+size 489763872