MistralForCausalLM

config.json

@@ -15,7 +15,7 @@
   "num_attention_heads": 32,
   "num_hidden_layers": 32,
   "num_key_value_heads": 8,
-  "rms_norm_eps": 1e-05,
+  "rms_norm_eps": 2e-06,
   "rope_theta": 10000.0,
   "sliding_window": 4096,
   "tie_word_embeddings": false,