feat(mistral): add Mistral model implementation and configs

- implement Mistral model in llm/models/mistral/mistral.py with GroupedQueryAttention, SwiGLU, RoPE, sliding window attention
- add __init__.py for module export
- add config files for mistral training and generation
- update universal experiment runner to support Mistral model
- add notebook for Mistral experiments

experiments/llm_only/configs/mistral_generate.json

@@ -0,0 +1,19 @@
+{
+    "bpe_tokenizer": "checkpoints/bpe_tokenizer.json",
+    "test_prompts": [
+      "Open weights",
+      "The Llama model is",
+      "Efficient transformers"
+    ],
+    "model_config_path": "checkpoints/mistral-bpe/config.json",
+    "model_weights": "checkpoints/mistral-bpe/model.pt",
+    "generation": {
+      "max_new_tokens": 40,
+      "temperature": 0.8,
+      "do_sample": true,
+      "top_k": null,
+      "top_p": null
+    },
+    "log_path": "checkpoints/mistral_only_generation_logs.json"
+  }
+  

experiments/llm_only/configs/mistral_train.json

@@ -0,0 +1,26 @@
+{
+    "bpe_tokenizer": "checkpoints/bpe_tokenizer.json",
+    "bpe_vocab_size": 1000,
+    "bpe_special_tokens": ["<pad>", "<unk>", "<bos>", "<eos>"],
+    "test_prompts": ["Open source AI", "What is Llama?"],
+    "model_config": {
+      "vocab_size": null,
+      "embed_dim": 256,
+      "num_q_heads": 4,
+      "num_kv_heads": 2,
+      "head_size": 64,
+      "num_layers": 4,
+      "max_position_embeddings": 512,
+      "window_size": 16,
+      "dropout": 0.1
+    },
+    "model_weights": "checkpoints/mistral-bpe/model.pt",
+    "model_config_path": "checkpoints/mistral-bpe/config.json",
+    "training": {
+      "learning_rate": 0.0003,
+      "batch_size": 2,
+      "num_epochs": 3,
+      "warmup_steps": 50
+    },
+    "log_path": "checkpoints/mistral_only_training_logs.json"
+  }

experiments/llm_only/run_llm_experiment.py

@@ -42,6 +42,9 @@ def load_model_class(model_name):
     elif model_name.lower() == 'llama':
         from llm.models.llama import Llama
         return Llama
+    elif model_name.lower() == 'mistral':
+        from llm.models.mistral import Mistral
+        return Mistral
     else:
         raise ValueError(f"Модель '{model_name}' не поддерживается.")