model update

README.md

@@ -0,0 +1,28 @@
+
+---
+widget:
+- text: "mammal is to whale what"
+  example_title: "Analogy Example 1 (semantic relation)"
+- text: "wedding is to marriage what "
+  example_title: "Analogy Example 2 (semantic relation, metaphor)"
+- text: "London is to U.K. what"
+  example_title: "Analogy Example 3 (entity)"
+- text: "actual is to actually what"
+  example_title: "Analogy Example 4 (morphological)"
+---
+# relbert/opt-125m-analogy
+
+This is [facebook/opt-125m](https://huggingface.co/facebook/opt-125m) fine-tuned on [relbert/semeval2012_relational_similarity](https://huggingface.co/datasets/relbert/semeval2012_relational_similarity) 
+for analogy generation, which is to generate a word pair (eg. `bird is to crow`) given a query (eg. `mammal is to whale`) 
+so that the query and the generated word pair form an analogy statement.  
+
+### Usage
+
+```python
+from transformers import pipeline
+
+pipe = pipeline('text2text-generation', model="relbert/opt-125m-analogy")
+output = pipe("mammal is to whale what")
+print(output)
+>>> [{'generated_text': 'bird is to crow'}]
+```

config.json

@@ -1,23 +1,17 @@
 {
-  "_name_or_path": "analogy_models/opt-125m-analogy-epoch10/model",
+  "_name_or_path": "facebook/opt-125m",
   "_remove_final_layer_norm": false,
   "activation_dropout": 0.0,
   "activation_function": "relu",
   "architectures": [
-    "T5ForConditionalGeneration"
+    "OPTForCausalLM"
   ],
   "attention_dropout": 0.0,
   "bos_token_id": 2,
-  "d_ff": 2048,
-  "d_kv": 64,
-  "d_model": 768,
-  "dense_act_fn": "relu",
   "do_layer_norm_before": true,
   "dropout": 0.1,
-  "dropout_rate": 0.1,
   "enable_bias": true,
   "eos_token_id": 2,
-  "feed_forward_proj": "relu",
   "ffn_dim": 3072,
   "finetuing_config": {
     "batch_size": 32,
@@ -29,22 +23,16 @@
     "random_seed": 42,
     "template": "<subj-a> is to <obj-a> what <subj-b> is to <obj-b>"
   },
+  "hidden_size": 768,
   "init_std": 0.02,
-  "initializer_factor": 1.0,
-  "is_encoder_decoder": true,
-  "is_gated_act": false,
   "layer_norm_elementwise_affine": true,
-  "layer_norm_epsilon": 1e-06,
   "layerdrop": 0.0,
   "max_position_embeddings": 2048,
-  "model_type": "t5",
-  "num_decoder_layers": 6,
-  "num_heads": 12,
-  "num_layers": 12,
+  "model_type": "opt",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
   "pad_token_id": 1,
   "prefix": "</s>",
-  "relative_attention_max_distance": 128,
-  "relative_attention_num_buckets": 32,
   "torch_dtype": "float32",
   "transformers_version": "4.26.1",
   "use_cache": true,

merges.txt

@@ -1,4 +1,4 @@
-#version: 0.2 - Trained by `huggingface/tokenizers`
+#version: 0.2
 Ġ t
 Ġ a
 h e

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa9225b427146a091257a647cf398db7c712f252a942515af23723ef0b3fa639
-size 607617881
+oid sha256:9b8e2bb320709a156e785d5c1b93e359a6a1b309c2cb6aa9eb2e0ee654899617
+size 501024733

tokenizer_config.json

@@ -19,7 +19,7 @@
   },
   "errors": "replace",
   "model_max_length": 1000000000000000019884624838656,
-  "name_or_path": "analogy_models/opt-125m-analogy-epoch10/model",
+  "name_or_path": "facebook/opt-125m",
   "pad_token": {
     "__type": "AddedToken",
     "content": "<pad>",

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07b9d163f9cd39545f020abdd398173acfcfc56dfe4a490404f260246ee31a7d
+size 3515

validation_accuracy.json

@@ -0,0 +1 @@
+{"accuracy": 0.5063291139240507, "dataset": "relbert/semeval2012_relational_similarity", "split": "validation"}