tiny random model

config.json

@@ -0,0 +1,33 @@
+{
+  "activation_dropout": 0.0,
+  "activation_function": "relu",
+  "architectures": [
+    "M2M100ForConditionalGeneration"
+  ],
+  "attention_dropout": 0.1,
+  "bos_token_id": 0,
+  "d_model": 16,
+  "decoder_attention_heads": 4,
+  "decoder_ffn_dim": 4,
+  "decoder_layerdrop": 0.05,
+  "decoder_layers": 2,
+  "decoder_start_token_id": 2,
+  "dropout": 0.1,
+  "encoder_attention_heads": 4,
+  "encoder_ffn_dim": 4,
+  "encoder_layerdrop": 0.05,
+  "encoder_layers": 2,
+  "eos_token_id": 2,
+  "gradient_checkpointing": false,
+  "init_std": 0.02,
+  "is_encoder_decoder": true,
+  "max_position_embeddings": 512,
+  "model_type": "m2m_100",
+  "num_hidden_layers": 2,
+  "pad_token_id": 1,
+  "scale_embedding": true,
+  "torch_dtype": "float16",
+  "transformers_version": "4.19.0.dev0",
+  "use_cache": true,
+  "vocab_size": 128112
+}

m2m-make-tiny-model.py

@@ -0,0 +1,56 @@
+#!/usr/bin/env python
+# coding: utf-8
+# Copyright 2020 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This script creates a super tiny model that is useful inside tests, when we just want to test that
+# the machinery works, without needing to the check the quality of the outcomes.
+#
+# This version is derived from https://huggingface.co/hf-internal-testing/tiny-random-m2m_100
+# but with max_position_embeddings=512 so that we don't need to recreate pos embeddings during forward
+#
+# It will be used then as "stas/tiny-m2m_100"
+
+# Build
+from transformers import M2M100Tokenizer, M2M100Config, M2M100ForConditionalGeneration
+
+mname = "hf-internal-testing/tiny-random-m2m_100"
+
+tokenizer = M2M100Tokenizer.from_pretrained(mname)
+# get the correct vocab sizes, etc. from the master model
+config = M2M100Config.from_pretrained(mname)
+# replicate the existing tiny model but we need longer max_position_embeddings
+config.update(dict(
+    max_position_embeddings=512,
+))
+
+tiny_model = M2M100ForConditionalGeneration(config)
+print(f"num of params {tiny_model.num_parameters()}")
+
+# Test
+model_inputs = tokenizer("Making tiny model", return_tensors="pt")
+gen_tokens = tiny_model.generate(**model_inputs, forced_bos_token_id=tokenizer.get_lang_id("fr"))
+print(tokenizer.batch_decode(gen_tokens, skip_special_tokens=True))
+#
+
+# Save
+mname_tiny = "tiny-m2m_100"
+tiny_model.half() # makes it smaller
+tiny_model.save_pretrained(mname_tiny)
+tokenizer.save_pretrained(mname_tiny)
+
+print(f"Generated {mname_tiny}")
+
+# Upload
+# transformers-cli upload tiny-m2m_100

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:62e77e2232830ad7d9bcf9bdbd1d447aa0bdaa3e0ed794b4ebda0d094c95e49e
+size 4140698

sentencepiece.bpe.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8f7c76ed2a5e0822be39f0a4f95a55eb19c78f4593ce609e2edbc2aea4d380a
+size 2423393

special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "<unk>", "sep_token": "</s>", "pad_token": "<pad>", "additional_special_tokens": ["__af__", "__am__", "__ar__", "__ast__", "__az__", "__ba__", "__be__", "__bg__", "__bn__", "__br__", "__bs__", "__ca__", "__ceb__", "__cs__", "__cy__", "__da__", "__de__", "__el__", "__en__", "__es__", "__et__", "__fa__", "__ff__", "__fi__", "__fr__", "__fy__", "__ga__", "__gd__", "__gl__", "__gu__", "__ha__", "__he__", "__hi__", "__hr__", "__ht__", "__hu__", "__hy__", "__id__", "__ig__", "__ilo__", "__is__", "__it__", "__ja__", "__jv__", "__ka__", "__kk__", "__km__", "__kn__", "__ko__", "__lb__", "__lg__", "__ln__", "__lo__", "__lt__", "__lv__", "__mg__", "__mk__", "__ml__", "__mn__", "__mr__", "__ms__", "__my__", "__ne__", "__nl__", "__no__", "__ns__", "__oc__", "__or__", "__pa__", "__pl__", "__ps__", "__pt__", "__ro__", "__ru__", "__sd__", "__si__", "__sk__", "__sl__", "__so__", "__sq__", "__sr__", "__ss__", "__su__", "__sv__", "__sw__", "__ta__", "__th__", "__tl__", "__tn__", "__tr__", "__uk__", "__ur__", "__uz__", "__vi__", "__wo__", "__xh__", "__yi__", "__yo__", "__zh__", "__zu__"]}

tokenizer_config.json

@@ -0,0 +1 @@
+{"src_lang": null, "tgt_lang": null, "bos_token": "<s>", "eos_token": "</s>", "sep_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>", "language_codes": "m2m100", "sp_model_kwargs": {}, "num_madeup_words": 8, "special_tokens_map_file": "m2m_100_1.2B_v2/special_tokens_map.json", "tokenizer_file": null, "name_or_path": "hf-internal-testing/tiny-random-m2m_100", "model_max_length": 1024, "additional_special_tokens": ["__af__", "__am__", "__ar__", "__ast__", "__az__", "__ba__", "__be__", "__bg__", "__bn__", "__br__", "__bs__", "__ca__", "__ceb__", "__cs__", "__cy__", "__da__", "__de__", "__el__", "__en__", "__es__", "__et__", "__fa__", "__ff__", "__fi__", "__fr__", "__fy__", "__ga__", "__gd__", "__gl__", "__gu__", "__ha__", "__he__", "__hi__", "__hr__", "__ht__", "__hu__", "__hy__", "__id__", "__ig__", "__ilo__", "__is__", "__it__", "__ja__", "__jv__", "__ka__", "__kk__", "__km__", "__kn__", "__ko__", "__lb__", "__lg__", "__ln__", "__lo__", "__lt__", "__lv__", "__mg__", "__mk__", "__ml__", "__mn__", "__mr__", "__ms__", "__my__", "__ne__", "__nl__", "__no__", "__ns__", "__oc__", "__or__", "__pa__", "__pl__", "__ps__", "__pt__", "__ro__", "__ru__", "__sd__", "__si__", "__sk__", "__sl__", "__so__", "__sq__", "__sr__", "__ss__", "__su__", "__sv__", "__sw__", "__ta__", "__th__", "__tl__", "__tn__", "__tr__", "__uk__", "__ur__", "__uz__", "__vi__", "__wo__", "__xh__", "__yi__", "__yo__", "__zh__", "__zu__"], "tokenizer_class": "M2M100Tokenizer"}