First version

app.py

@@ -0,0 +1,79 @@
+import gradio as gr
+import transformers as tr
+from src.models.fine_tune_gpt2 import SPECIAL_TOKENS
+
+MPATH = "./mdl_roomgen2"
+MODEL = tr.GPT2LMHeadModel.from_pretrained(MPATH)
+
+# ToDo: Will save tokenizer next time so can replace this with a load
+TOK = tr.GPT2Tokenizer.from_pretrained("gpt2")
+TOK.add_special_tokens(SPECIAL_TOKENS)
+
+
+def generate_room(room_name, method, max_length):
+    """
+    Uses pretrained model to generate text for a dungeon room
+
+    Args:
+        room_name:
+        method:
+
+    Returns:
+
+    """
+    prompt = " ".join(
+        [
+            SPECIAL_TOKENS["bos_token"],
+            room_name,
+            SPECIAL_TOKENS["sep_token"]
+        ]
+    )
+    ids = TOK.encode(prompt, return_tensors="pt")
+    if method == "Greedy":
+        output = MODEL.generate(ids, max_length=max_length)
+    elif method == "Beam":
+        output = MODEL.generate(
+            ids,
+            max_length=max_length,
+            num_beams=5,
+            no_repeat_ngram_size=4,
+            early_stopping=True
+        )
+    elif method == "Sample":
+        # Sample
+        output = MODEL.generate(
+            ids,
+            max_length=max_length,
+            do_sample=True,
+            top_k=0
+        )
+    elif method == "Top K":
+        # Top K - redistribute probability over top words, see Fan et al. 2018
+        output = MODEL.generate(
+            ids,
+            max_length=max_length,
+            do_sample=True,
+            top_k=50
+        )
+    else:
+        raise ValueError(f"Unexpected generation method, received {method}")
+    output = TOK.decode(output[0][ids.shape[1]:], clean_up_tokenization_spaces=True).replace("  ", " ")
+    return output
+
+
+if __name__ == "__main__":
+    iface = gr.Interface(
+        title="RPG Room Generator",
+        fn=generate_room,
+        inputs=[
+            gr.inputs.Textbox(lines=1),
+            gr.inputs.Radio(["Greedy", "Beam", "Sample", "Top K"], default="Top K"),
+            gr.inputs.Slider(minimum=50, maximum=250, default=100)
+        ],
+        outputs="text",
+        examples=[["Dark Catacombs", "Top K", 150]],
+        layout="horizontal",
+        allow_flagging=None,
+        theme="dark"
+    )
+    app, local_url, share_url = iface.launch()

mdl_roomgen2/config.json

@@ -0,0 +1,41 @@
+{
+  "_name_or_path": "gpt2",
+  "activation_function": "gelu_new",
+  "architectures": [
+    "GPT2LMHeadModel"
+  ],
+  "attn_pdrop": 0.1,
+  "bos_token_id": 50256,
+  "embd_pdrop": 0.1,
+  "eos_token_id": 50257,
+  "initializer_range": 0.02,
+  "layer_norm_epsilon": 1e-05,
+  "model_type": "gpt2",
+  "n_ctx": 1024,
+  "n_embd": 768,
+  "n_head": 12,
+  "n_inner": null,
+  "n_layer": 12,
+  "n_positions": 1024,
+  "pad_token_id": 50258,
+  "reorder_and_upcast_attn": false,
+  "resid_pdrop": 0.1,
+  "scale_attn_by_inverse_layer_idx": false,
+  "scale_attn_weights": true,
+  "sep_token_id": 50259,
+  "summary_activation": null,
+  "summary_first_dropout": 0.1,
+  "summary_proj_to_labels": true,
+  "summary_type": "cls_index",
+  "summary_use_proj": true,
+  "task_specific_params": {
+    "text-generation": {
+      "do_sample": true,
+      "max_length": 50
+    }
+  },
+  "torch_dtype": "float32",
+  "transformers_version": "4.17.0",
+  "use_cache": true,
+  "vocab_size": 50260
+}

mdl_roomgen2/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:82f9411bb82b07718b5c1dcf60326fde7e2ba639d880f96b05e0a6db58e06b3d
+size 510413609

requirements.txt

@@ -0,0 +1,10 @@
+datasets >= 1.8.0
+transformers >= 4.17.0
+bs4~=4.10.0
+pandas~=1.4.1
+requests~=2.27.1
+beautifulsoup4~=4.8.2
+tqdm~=4.63.0
+wandb~=0.12.13
+scikit-learn~=1.0.2
+gradio~=2.9.1