chatglm中增加generation_config

chatglm-6b-int4/bert4torch_config.json

@@ -21,5 +21,6 @@
     "tie_emb_prj_weight": false,
     "quantization_bit": 4,
     "quantization_method": "cpm_kernels",
-    "target_modules": ["q", "k", "v", "o", "intermediateDense", "outputDense"]
+    "target_modules": ["q", "k", "v", "o", "intermediateDense", "outputDense"],
+    "generation_config": {"max_length": 2048}
 }

chatglm-6b-int8/bert4torch_config.json

@@ -21,5 +21,6 @@
     "quantization_bit": 8,
     "quantization_method": "cpm_kernels",
     "target_modules": ["q", "k", "v", "o", "intermediateDense", "outputDense"],
-    "tie_emb_prj_weight": false
+    "tie_emb_prj_weight": false,
+    "generation_config": {"max_length": 2048}
     }

chatglm-6b-v0.1.0/bert4torch_config.json

@@ -18,5 +18,6 @@
     "segment_vocab_size": 0,
     "skip_init": true,
     "rope_rank": "updown",
-    "tie_emb_prj_weight": false
+    "tie_emb_prj_weight": false,
+    "generation_config": {"max_length": 2048}
 }

chatglm-6b/bert4torch_config.json

@@ -18,5 +18,6 @@
     "segment_vocab_size": 0,
     "skip_init": true,
     "rope_rank": "updown",
-    "tie_emb_prj_weight": false
+    "tie_emb_prj_weight": false,
+    "generation_config": {"max_length": 2048}
 }

chatglm2-6b-32k/bert4torch_config.json

@@ -18,5 +18,6 @@
   "rope_rank": "adjacent",
   "rope_ratio": 16,
   "position_encoding_2d": true,
-  "flash_attention": true
+  "flash_attention": true,
+  "generation_config": {"max_length": 32768}
 }

chatglm2-6b-int4/bert4torch_config.json

@@ -20,5 +20,6 @@
     "flash_attention": true,
     "quantization_bit": 4,
     "quantization_method": "cpm_kernels",
-    "target_modules": ["q", "k", "v", "o", "intermediateDense", "outputDense"]
+    "target_modules": ["q", "k", "v", "o", "intermediateDense", "outputDense"],
+    "generation_config": {"max_length": 32768}
     }

chatglm2-6b/bert4torch_config.json

@@ -17,5 +17,6 @@
     "rmsnorm": true,
     "rope_rank": "adjacent",
     "position_encoding_2d": true,
-    "flash_attention": true
+    "flash_attention": true,
+    "generation_config": {"max_length": 32768}
     }

chatglm3-6b-32k/bert4torch_config.json

@@ -18,5 +18,6 @@
   "rope_rank": "adjacent",
   "ntk_alpha": 44.24632364389211,
   "position_encoding_2d": true,
-  "flash_attention": true
+  "flash_attention": true,
+  "generation_config": {"end_id": [2, 64795, 64797], "max_length": 32768}
 }

chatglm3-6b/bert4torch_config.json

@@ -17,5 +17,6 @@
     "rmsnorm": true,
     "rope_rank": "adjacent",
     "position_encoding_2d": true,
-    "flash_attention": true
+    "flash_attention": true,
+    "generation_config": {"end_id": [2, 64795, 64797], "max_length": 8192}
 }

internlm-chat-7b/bert4torch_config.json

@@ -12,5 +12,6 @@
     "skip_init": true,
     "rope_rank": "updown",
     "torch_dtype": "float16",
-    "tie_word_embeddings": false
+    "tie_word_embeddings": false,
+    "generation_config": {"tokenizer_config": {"skip_special_tokens": true}, "end_id": [2, 103028]}
 }