增加Qwen3

.gitignore

@@ -1,9 +1,7 @@
 # 忽略所有文件
 *
-config.json
-generation_config.json
-Tongjilibo
-
-# 允许 .json 文件
 !bert4torch_config.json
-
+!*/
+Tongjilibo
+__tensorflow_weights
+maidalun1020

Qwen/Qwen3-0.6B/bert4torch_config.json

@@ -0,0 +1,30 @@
+{
+    "model": "qwen3",
+    "hidden_act": "silu",
+	"attention_probs_dropout_prob": 0.0,
+    "bos_token_id": 151643,
+    "eos_token_id": 151645,
+    "attention_head_size": 128,
+    "attention_key_size": 128,
+    "intermediate_size": 3072,
+    "initializer_range": 0.02,
+    "layer_norm_eps": 1e-06,
+    "hidden_size": 1024,
+    "num_attention_heads": 16,
+    "num_hidden_layers": 28,
+	"num_key_value_heads": 8,
+    "rope_theta": 1000000.0,
+    "tie_word_embeddings": true,
+	"torch_dtype": "bfloat16",
+    "_attn_implementation": "sdpa",
+    "vocab_size": 151936,
+    "segment_vocab_size": 0,
+    "skip_init": true,
+    "rope_rank": "updown",
+	"max_position_embeddings": 40960,
+	"sliding_window": null,
+	"max_window_layers": 28,
+    "convert_lm_logits_dtype": "float32",
+    "generation_config": {"tokenizer_config": {"skip_special_tokens": true}, "eos_token_id": [151643, 151645],
+                          "max_length": 40960}
+}

Qwen/Qwen3-1.7B/bert4torch_config.json

@@ -0,0 +1,30 @@
+{
+    "model": "qwen3",
+    "hidden_act": "silu",
+	"attention_probs_dropout_prob": 0.0,
+    "bos_token_id": 151643,
+    "eos_token_id": 151645,
+    "attention_head_size": 128,
+    "attention_key_size": 128,
+    "intermediate_size": 6144,
+    "initializer_range": 0.02,
+    "layer_norm_eps": 1e-06,
+    "hidden_size": 2048,
+    "num_attention_heads": 16,
+    "num_hidden_layers": 28,
+	"num_key_value_heads": 8,
+    "rope_theta": 1000000.0,
+    "tie_word_embeddings": true,
+	"torch_dtype": "bfloat16",
+    "_attn_implementation": "sdpa",
+    "vocab_size": 151936,
+    "segment_vocab_size": 0,
+    "skip_init": true,
+    "rope_rank": "updown",
+	"max_position_embeddings": 40960,
+	"sliding_window": null,
+	"max_window_layers": 28,
+    "convert_lm_logits_dtype": "float32",
+    "generation_config": {"tokenizer_config": {"skip_special_tokens": true}, "eos_token_id": [151643, 151645],
+                          "max_length": 40960}
+}

Qwen/Qwen3-14B/bert4torch_config.json

@@ -0,0 +1,30 @@
+{
+    "model": "qwen3",
+    "hidden_act": "silu",
+	"attention_probs_dropout_prob": 0.0,
+    "bos_token_id": 151643,
+    "eos_token_id": 151645,
+    "attention_head_size": 128,
+    "attention_key_size": 128,
+    "intermediate_size": 17408,
+    "initializer_range": 0.02,
+    "layer_norm_eps": 1e-06,
+    "hidden_size": 5120,
+    "num_attention_heads": 40,
+    "num_hidden_layers": 40,
+	"num_key_value_heads": 8,
+    "rope_theta": 1000000.0,
+    "tie_word_embeddings": false,
+	"torch_dtype": "bfloat16",
+    "_attn_implementation": "sdpa",
+    "vocab_size": 151936,
+    "segment_vocab_size": 0,
+    "skip_init": true,
+    "rope_rank": "updown",
+	"max_position_embeddings": 40960,
+	"sliding_window": null,
+	"max_window_layers": 40,
+    "convert_lm_logits_dtype": "float32",
+    "generation_config": {"tokenizer_config": {"skip_special_tokens": true}, "eos_token_id": [151643, 151645],
+                          "max_length": 40960}
+}

Qwen/Qwen3-32B/bert4torch_config.json

@@ -0,0 +1,30 @@
+{
+    "model": "qwen3",
+    "hidden_act": "silu",
+	"attention_probs_dropout_prob": 0.0,
+    "bos_token_id": 151643,
+    "eos_token_id": 151645,
+    "attention_head_size": 128,
+    "attention_key_size": 128,
+    "intermediate_size": 25600,
+    "initializer_range": 0.02,
+    "layer_norm_eps": 1e-06,
+    "hidden_size": 5120,
+    "num_attention_heads": 64,
+    "num_hidden_layers": 64,
+	"num_key_value_heads": 8,
+    "rope_theta": 1000000.0,
+    "tie_word_embeddings": false,
+	"torch_dtype": "bfloat16",
+    "_attn_implementation": "sdpa",
+    "vocab_size": 151936,
+    "segment_vocab_size": 0,
+    "skip_init": true,
+    "rope_rank": "updown",
+	"max_position_embeddings": 40960,
+	"sliding_window": null,
+	"max_window_layers": 64,
+    "convert_lm_logits_dtype": "float32",
+    "generation_config": {"tokenizer_config": {"skip_special_tokens": true}, "eos_token_id": [151643, 151645],
+                          "max_length": 40960}
+}

Qwen/Qwen3-4B/bert4torch_config.json

@@ -0,0 +1,30 @@
+{
+    "model": "qwen3",
+    "hidden_act": "silu",
+	"attention_probs_dropout_prob": 0.0,
+    "bos_token_id": 151643,
+    "eos_token_id": 151645,
+    "attention_head_size": 128,
+    "attention_key_size": 128,
+    "intermediate_size": 9728,
+    "initializer_range": 0.02,
+    "layer_norm_eps": 1e-06,
+    "hidden_size": 2560,
+    "num_attention_heads": 32,
+    "num_hidden_layers": 36,
+	"num_key_value_heads": 8,
+    "rope_theta": 1000000.0,
+    "tie_word_embeddings": true,
+	"torch_dtype": "bfloat16",
+    "_attn_implementation": "sdpa",
+    "vocab_size": 151936,
+    "segment_vocab_size": 0,
+    "skip_init": true,
+    "rope_rank": "updown",
+	"max_position_embeddings": 40960,
+	"sliding_window": null,
+	"max_window_layers": 36,
+    "convert_lm_logits_dtype": "float32",
+    "generation_config": {"tokenizer_config": {"skip_special_tokens": true}, "eos_token_id": [151643, 151645],
+                          "max_length": 40960}
+}

Qwen/Qwen3-8B/bert4torch_config.json

@@ -0,0 +1,30 @@
+{
+    "model": "qwen3",
+    "hidden_act": "silu",
+	"attention_probs_dropout_prob": 0.0,
+    "bos_token_id": 151643,
+    "eos_token_id": 151645,
+    "attention_head_size": 128,
+    "attention_key_size": 128,
+    "intermediate_size": 12288,
+    "initializer_range": 0.02,
+    "layer_norm_eps": 1e-06,
+    "hidden_size": 4096,
+    "num_attention_heads": 32,
+    "num_hidden_layers": 36,
+	"num_key_value_heads": 8,
+    "rope_theta": 1000000.0,
+    "tie_word_embeddings": false,
+	"torch_dtype": "bfloat16",
+    "_attn_implementation": "sdpa",
+    "vocab_size": 151936,
+    "segment_vocab_size": 0,
+    "skip_init": true,
+    "rope_rank": "updown",
+	"max_position_embeddings": 40960,
+	"sliding_window": null,
+	"max_window_layers": 36,
+    "convert_lm_logits_dtype": "float32",
+    "generation_config": {"tokenizer_config": {"skip_special_tokens": true}, "eos_token_id": [151643, 151645],
+                          "max_length": 40960}
+}