Upload V2: stackformer GPT-2 + sparse cross-attention vision model (128 visual tokens)

config.json

@@ -0,0 +1,25 @@
+{
+  "vocab_size": 50258,
+  "embed_dim": 768,
+  "num_layers": 12,
+  "num_heads": 12,
+  "hidden_dim": 3072,
+  "context_length": 128,
+  "dropout": 0.1,
+  "qkv_bias": true,
+  "cross_attention_pos": [
+    3,
+    6,
+    9
+  ],
+  "num_visual_tokens": 128,
+  "vision_dim": 768,
+  "perceiver_depth": 2,
+  "perceiver_heads": 8,
+  "batch_size": 16,
+  "grad_accum_steps": 2,
+  "learning_rate": 0.0002,
+  "weight_decay": 0.01,
+  "time_budget_minutes": 270,
+  "amp_dtype": "bf16"
+}

model_checkpoint.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b734ff8caff36c441a4ae58e20fdfafbde1b5b7ff4a2c5d47c2c9370db87bc9
+size 1062433014

tokenizer/added_tokens.json

@@ -0,0 +1,3 @@
+{
+  "<|pad|>": 50257
+}

tokenizer/special_tokens_map.json

@@ -0,0 +1,12 @@
+{
+  "bos_token": "<|endoftext|>",
+  "eos_token": "<|endoftext|>",
+  "pad_token": {
+    "content": "<|pad|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": "<|endoftext|>"
+}

tokenizer/tokenizer_config.json

@@ -0,0 +1,28 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "50256": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "50257": {
+      "content": "<|pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<|endoftext|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|endoftext|>",
+  "model_max_length": 1024,
+  "pad_token": "<|pad|>",
+  "tokenizer_class": "GPT2Tokenizer",
+  "unk_token": "<|endoftext|>"
+}