Upload folder using huggingface_hub

config.json

@@ -1,32 +1,50 @@
 {
-  "_name_or_path": "microsoft/phi-2",
+  "_name_or_path": "cognitivecomputations/dolphin-2_6-phi-2",
   "activation_function": "gelu_new",
   "architectures": [
-    "PhiForCausalLM"
+    "MixtralForCausalLM"
   ],
+  "attention_dropout": 0.0,
   "attn_pdrop": 0.0,
   "auto_map": {
-    "AutoConfig": "configuration_phi.PhiConfig",
-    "AutoModelForCausalLM": "modeling_phi.PhiForCausalLM"
+    "AutoConfig": "cognitivecomputations/dolphin-2_6-phi-2--configuration_phi.PhiConfig",
+    "AutoModelForCausalLM": "cognitivecomputations/dolphin-2_6-phi-2--modeling_phi.PhiForCausalLM"
   },
+  "bos_token_id": null,
   "embd_pdrop": 0.0,
+  "eos_token_id": null,
   "flash_attn": false,
   "flash_rotary": false,
   "fused_dense": false,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
   "img_processor": null,
   "initializer_range": 0.02,
+  "intermediate_size": 14336,
   "layer_norm_epsilon": 1e-05,
-  "model_type": "phi-msft",
+  "max_position_embeddings": 2048,
+  "model_type": "mixtral",
   "n_embd": 2560,
   "n_head": 32,
   "n_head_kv": null,
   "n_inner": null,
   "n_layer": 32,
   "n_positions": 2048,
+  "num_attention_heads": 32,
+  "num_experts_per_tok": 2,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 8,
+  "num_local_experts": 2,
+  "output_router_logits": false,
   "resid_pdrop": 0.1,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000.0,
   "rotary_dim": 32,
+  "router_aux_loss_coef": 0.001,
+  "sliding_window": null,
   "tie_word_embeddings": false,
   "torch_dtype": "float16",
-  "transformers_version": "4.35.2",
+  "transformers_version": "4.36.2",
+  "use_cache": false,
   "vocab_size": 51200
 }

mergekit_moe_config.yml

@@ -4,4 +4,4 @@ experts:
   - source_model: cognitivecomputations/dolphin-2_6-phi-2
     positive_prompts: [""]
   - source_model: lxuechen/phi-2-dpo
-    positive_prompts: [""]
+    positive_prompts: [""]

modeling_phi.py

@@ -294,7 +294,7 @@ class MoE(nn.Module):
     def __init__(
         self,
         config: PretrainedConfig,
-        num_experts=4,
+        num_experts=2,
         num_experts_per_tok=2,
         num_shards=1,
         **kwargs,