Add correct Mixtral config.json for lm-eval

config.json

@@ -0,0 +1,18 @@
+{
+  "architectures": [
+    "MixtralForCausalLM"
+  ],
+  "model_type": "mixtral",
+  "hidden_size": 4096,
+  "intermediate_size": 14336,
+  "max_position_embeddings": 32768,
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 8,
+  "num_local_experts": 8,
+  "num_experts_per_tok": 2,
+  "rms_norm_eps": 1e-05,
+  "rope_theta": 1000000.0,
+  "torch_dtype": "bfloat16",
+  "vocab_size": 32000
+}