Add llm_config.json (Q4+int4 PLE, requires TokForge 3.4.9)

llm_config.json

@@ -0,0 +1,48 @@
+{
+    "model_type": "gemma4",
+    "hidden_size": 2560,
+    "attention_mask": "float",
+    "attention_type": "mix",
+    "is_mrope": false,
+    "layer_nums": 42,
+    "hidden_size_per_layer_input": 256,
+    "per_layer_embedding_hidden_size": 10752,
+    "per_layer_embedding_file": "per_layer_embeddings_int4.bin",
+    "per_layer_scale_emb": 16.0,
+    "use_template": true,
+    "force_full_decode_recompute": false,
+    "jinja": {
+        "chat_template": "{{ bos_token }}{% for message in messages %}{% if message.role == \"system\" %}<|turn>system\n{{ message.content }}<turn|>\n{% elif message.role == \"user\" %}<|turn>user\n{{ message.content }}<turn|>\n{% elif message.role == \"assistant\" %}<|turn>model\n{{ message.content }}<turn|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|turn>model\n{% endif %}",
+        "bos": "<bos>",
+        "eos": "<turn|>"
+    },
+    "scale_emb": 50.5,
+    "sampler_type": "mixed",
+    "temperature": 1.0,
+    "top_k": 64,
+    "top_p": 0.95,
+    "min_p": 0.0,
+    "tfs_z": 1.0,
+    "typical": 1.0,
+    "repetition_penalty": 1.0,
+    "presence_penalty": 0.0,
+    "frequency_penalty": 0.0,
+    "penalty_window": 0,
+    "n_gram": 8,
+    "ngram_factor": 1.0,
+    "sliding_window": 512,
+    "tie_embeddings": [
+        0,
+        335544320,
+        41943040,
+        4,
+        128
+    ],
+    "ple_quant": [
+        0,
+        1409286144,
+        176160768,
+        4,
+        128
+    ]
+}