Switch export_model to use AutoModel and AutoTokenizer (#1260)

* refactor export_model to use AutoModel and AutoTokenizer * use AutoConfig, AutoTokenizer, and AutoModel instead of jiant model_type * Switch to hf_pretrained_model_name_or_path. Remove unused tokenizer_path. Update notebooks with AutoClass changes.
nyu-mll · Jan 19, 2021 · 723786a · 723786a
1 parent 6e6c2e3
commit 723786a
Show file tree

Hide file tree

Showing 21 changed files with 150 additions and 249 deletions.
diff --git a/examples/notebooks/jiant_Basic_Example.ipynb b/examples/notebooks/jiant_Basic_Example.ipynb
@@ -158,8 +158,8 @@
    },
    "outputs": [],
    "source": [
-    "export_model.lookup_and_export_model(\n",
-    "    model_type=\"roberta-base\",\n",
+    "export_model.export_model(\n",
+    "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "    output_base_path=\"./models/roberta-base\",\n",
     ")"
    ]
@@ -191,8 +191,7 @@
     "\n",
     "tokenize_and_cache.main(tokenize_and_cache.RunConfiguration(\n",
     "    task_config_path=f\"./tasks/configs/{task_name}_config.json\",\n",
-    "    model_type=\"roberta-base\",\n",
-    "    model_tokenizer_path=\"./models/roberta-base/tokenizer\",\n",
+    "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "    output_dir=f\"./cache/{task_name}\",\n",
     "    phases=[\"train\", \"val\"],\n",
     "))"
@@ -309,10 +308,9 @@
     "run_args = main_runscript.RunConfiguration(\n",
     "    jiant_task_container_config_path=\"./run_configs/mrpc_run_config.json\",\n",
     "    output_dir=\"./runs/mrpc\",\n",
-    "    model_type=\"roberta-base\",\n",
+    "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "    model_path=\"./models/roberta-base/model/roberta-base.p\",\n",
     "    model_config_path=\"./models/roberta-base/model/roberta-base.json\",\n",
-    "    model_tokenizer_path=\"./models/roberta-base/tokenizer\",\n",
     "    learning_rate=1e-5,\n",
     "    eval_every_steps=500,\n",
     "    do_train=True,\n",

diff --git a/examples/notebooks/jiant_EdgeProbing_Example.ipynb b/examples/notebooks/jiant_EdgeProbing_Example.ipynb
@@ -2702,8 +2702,8 @@
         "outputId": "c21bdffa-0ff3-49f3-e734-af5530ab4711"
       },
       "source": [
-        "export_model.lookup_and_export_model(\n",
-        "    model_type=\"roberta-base\",\n",
+        "export_model.export_model(\n",
+        "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
         "    output_base_path=\"./models/roberta-base\",\n",
         ")"
       ],
@@ -2856,8 +2856,7 @@
         "\n",
         "tokenize_and_cache.main(tokenize_and_cache.RunConfiguration(\n",
         "    task_config_path=f\"./tasks/configs/{task_name}_config.json\",\n",
-        "    model_type=\"roberta-base\",\n",
-        "    model_tokenizer_path=\"./models/roberta-base/tokenizer\",\n",
+        "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
         "    output_dir=f\"./cache/{task_name}\",\n",
         "    phases=[\"train\", \"val\"],\n",
         "))"
@@ -3147,10 +3146,9 @@
         "run_args = main_runscript.RunConfiguration(\n",
         "    jiant_task_container_config_path=\"./run_configs/semeval_run_config.json\",\n",
         "    output_dir=\"./runs/semeval\",\n",
-        "    model_type=\"roberta-base\",\n",
+        "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
         "    model_path=\"./models/roberta-base/model/roberta-base.p\",\n",
         "    model_config_path=\"./models/roberta-base/model/roberta-base.json\",\n",
-        "    model_tokenizer_path=\"./models/roberta-base/tokenizer\",\n",
         "    learning_rate=1e-5,\n",
         "    eval_every_steps=500,\n",
         "    do_train=True,\n",
@@ -3170,7 +3168,6 @@
             "  model_type: roberta-base\n",
             "  model_path: ./models/roberta-base/model/roberta-base.p\n",
             "  model_config_path: ./models/roberta-base/model/roberta-base.json\n",
-            "  model_tokenizer_path: ./models/roberta-base/tokenizer\n",
             "  model_load_mode: from_transformers\n",
             "  do_train: True\n",
             "  do_val: True\n",
@@ -3204,7 +3201,6 @@
             "  \"model_type\": \"roberta-base\",\n",
             "  \"model_path\": \"./models/roberta-base/model/roberta-base.p\",\n",
             "  \"model_config_path\": \"./models/roberta-base/model/roberta-base.json\",\n",
-            "  \"model_tokenizer_path\": \"./models/roberta-base/tokenizer\",\n",
             "  \"model_load_mode\": \"from_transformers\",\n",
             "  \"do_train\": true,\n",
             "  \"do_val\": true,\n",

diff --git a/examples/notebooks/jiant_MNLI_Diagnostic_Example.ipynb b/examples/notebooks/jiant_MNLI_Diagnostic_Example.ipynb
@@ -140,8 +140,8 @@
    },
    "outputs": [],
    "source": [
-    "export_model.lookup_and_export_model(\n",
-    "    model_type=\"roberta-base\",\n",
+    "export_model.export_model(\n",
+    "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "    output_base_path=\"./models/roberta-base\",\n",
     ")"
    ]
@@ -169,24 +169,21 @@
     "# Tokenize and cache each task\n",
     "tokenize_and_cache.main(tokenize_and_cache.RunConfiguration(\n",
     "    task_config_path=f\"./tasks/configs/mnli_config.json\",\n",
-    "    model_type=\"roberta-base\",\n",
-    "    model_tokenizer_path=\"./models/roberta-base/tokenizer\",\n",
+    "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "    output_dir=f\"./cache/mnli\",\n",
     "    phases=[\"train\", \"val\"],\n",
     "))\n",
     "\n",
     "tokenize_and_cache.main(tokenize_and_cache.RunConfiguration(\n",
     "    task_config_path=f\"./tasks/configs/mnli_mismatched_config.json\",\n",
-    "    model_type=\"roberta-base\",\n",
-    "    model_tokenizer_path=\"./models/roberta-base/tokenizer\",\n",
+    "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "    output_dir=f\"./cache/mnli_mismatched\",\n",
     "    phases=[\"val\"],\n",
     "))\n",
     "\n",
     "tokenize_and_cache.main(tokenize_and_cache.RunConfiguration(\n",
     "    task_config_path=f\"./tasks/configs/glue_diagnostics_config.json\",\n",
-    "    model_type=\"roberta-base\",\n",
-    "    model_tokenizer_path=\"./models/roberta-base/tokenizer\",\n",
+    "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "    output_dir=f\"./cache/glue_diagnostics\",\n",
     "    phases=[\"test\"],\n",
     "))"
@@ -323,10 +320,9 @@
     "run_args = main_runscript.RunConfiguration(\n",
     "    jiant_task_container_config_path=\"./run_configs/jiant_run_config.json\",\n",
     "    output_dir=\"./runs/run1\",\n",
-    "    model_type=\"roberta-base\",\n",
+    "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "    model_path=\"./models/roberta-base/model/roberta-base.p\",\n",
     "    model_config_path=\"./models/roberta-base/model/roberta-base.json\",\n",
-    "    model_tokenizer_path=\"./models/roberta-base/tokenizer\",\n",
     "    learning_rate=1e-5,\n",
     "    eval_every_steps=500,\n",
     "    do_train=True,\n",

diff --git a/examples/notebooks/jiant_Multi_Task_Example.ipynb b/examples/notebooks/jiant_Multi_Task_Example.ipynb
@@ -161,8 +161,8 @@
    },
    "outputs": [],
    "source": [
-    "export_model.lookup_and_export_model(\n",
-    "    model_type=\"roberta-base\",\n",
+    "export_model.export_model(\n",
+    "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "    output_base_path=\"./models/roberta-base\",\n",
     ")"
    ]
@@ -193,8 +193,7 @@
     "for task_name in [\"rte\", \"stsb\", \"commonsenseqa\"]:\n",
     "    tokenize_and_cache.main(tokenize_and_cache.RunConfiguration(\n",
     "        task_config_path=f\"./tasks/configs/{task_name}_config.json\",\n",
-    "        model_type=\"roberta-base\",\n",
-    "        model_tokenizer_path=\"./models/roberta-base/tokenizer\",\n",
+    "        hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "        output_dir=f\"./cache/{task_name}\",\n",
     "        phases=[\"train\", \"val\"],\n",
     "    ))"
@@ -342,10 +341,9 @@
     "run_args = main_runscript.RunConfiguration(\n",
     "    jiant_task_container_config_path=\"./run_configs/jiant_run_config.json\",\n",
     "    output_dir=\"./runs/run1\",\n",
-    "    model_type=\"roberta-base\",\n",
+    "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "    model_path=\"./models/roberta-base/model/roberta-base.p\",\n",
     "    model_config_path=\"./models/roberta-base/model/roberta-base.json\",\n",
-    "    model_tokenizer_path=\"./models/roberta-base/tokenizer\",\n",
     "    learning_rate=1e-5,\n",
     "    eval_every_steps=500,\n",
     "    do_train=True,\n",

diff --git a/examples/notebooks/jiant_STILTs_Example.ipynb b/examples/notebooks/jiant_STILTs_Example.ipynb
@@ -163,8 +163,8 @@
    },
    "outputs": [],
    "source": [
-    "export_model.lookup_and_export_model(\n",
-    "    model_type=\"roberta-base\",\n",
+    "export_model.export_model(\n",
+    "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "    output_base_path=\"./models/roberta-base\",\n",
     ")"
    ]
@@ -195,8 +195,7 @@
     "for task_name in [\"mnli\", \"rte\"]:\n",
     "    tokenize_and_cache.main(tokenize_and_cache.RunConfiguration(\n",
     "        task_config_path=f\"./tasks/configs/{task_name}_config.json\",\n",
-    "        model_type=\"roberta-base\",\n",
-    "        model_tokenizer_path=\"./models/roberta-base/tokenizer\",\n",
+    "        hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "        output_dir=f\"./cache/{task_name}\",\n",
     "        phases=[\"train\", \"val\"],\n",
     "    ))"
@@ -367,10 +366,9 @@
     "run_args = main_runscript.RunConfiguration(\n",
     "    jiant_task_container_config_path=\"./run_configs/mnli_run_config.json\",\n",
     "    output_dir=\"./runs/mnli\",\n",
-    "    model_type=\"roberta-base\",\n",
+    "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "    model_path=\"./models/roberta-base/model/roberta-base.p\",\n",
     "    model_config_path=\"./models/roberta-base/model/roberta-base.json\",\n",
-    "    model_tokenizer_path=\"./models/roberta-base/tokenizer\",\n",
     "    learning_rate=1e-5,\n",
     "    eval_every_steps=500,\n",
     "    do_train=True,\n",
@@ -404,11 +402,10 @@
     "run_args = main_runscript.RunConfiguration(\n",
     "    jiant_task_container_config_path=\"./run_configs/rte_run_config.json\",\n",
     "    output_dir=\"./runs/mnli___rte\",\n",
-    "    model_type=\"roberta-base\",\n",
+    "    hf_pretrained_model_name_or_path=\"roberta-base\",\n",
     "    model_path=\"./runs/mnli/best_model.p\",  # Loading the best model\n",
     "    model_load_mode=\"partial\",\n",
     "    model_config_path=\"./models/roberta-base/model/roberta-base.json\",\n",
-    "    model_tokenizer_path=\"./models/roberta-base/tokenizer\",\n",
     "    learning_rate=1e-5,\n",
     "    eval_every_steps=500,\n",
     "    do_train=True,\n",

diff --git a/examples/notebooks/jiant_XNLI_Example.ipynb b/examples/notebooks/jiant_XNLI_Example.ipynb
@@ -164,8 +164,8 @@
    },
    "outputs": [],
    "source": [
-    "export_model.lookup_and_export_model(\n",
-    "    model_type=\"xlm-roberta-base\",\n",
+    "export_model.export_model(\n",
+    "    hf_pretrained_model_name_or_path=\"xlm-roberta-base\",\n",
     "    output_base_path=\"./models/xlm-roberta-base\",\n",
     ")"
    ]
@@ -197,8 +197,7 @@
     "# Tokenize and cache MNLI\n",
     "tokenize_and_cache.main(tokenize_and_cache.RunConfiguration(\n",
     "    task_config_path=f\"./tasks/configs/mnli_config.json\",\n",
-    "    model_type=\"xlm-roberta-base\",\n",
-    "    model_tokenizer_path=\"./models/xlm-roberta-base/tokenizer\",\n",
+    "    hf_pretrained_model_name_or_path=\"xlm-roberta-base\",\n",
     "    output_dir=f\"./cache/mnli\",\n",
     "    phases=[\"train\", \"val\"],\n",
     "))\n",
@@ -207,8 +206,7 @@
     "for lang in [\"de\", \"zh\"]:\n",
     "    tokenize_and_cache.main(tokenize_and_cache.RunConfiguration(\n",
     "        task_config_path=f\"./tasks/configs/xnli_{lang}_config.json\",\n",
-    "        model_type=\"xlm-roberta-base\",\n",
-    "        model_tokenizer_path=\"./models/xlm-roberta-base/tokenizer\",\n",
+    "        hf_pretrained_model_name_or_path=\"xlm-roberta-base\",\n",
     "        output_dir=f\"./cache/xnli_{lang}\",\n",
     "        phases=[\"val\"],\n",
     "    ))"
@@ -384,10 +382,9 @@
     "run_args = main_runscript.RunConfiguration(\n",
     "    jiant_task_container_config_path=\"./run_configs/jiant_run_config.json\",\n",
     "    output_dir=\"./runs/run1\",\n",
-    "    model_type=\"xlm-roberta-base\",\n",
+    "    hf_pretrained_model_name_or_path=\"xlm-roberta-base\",\n",
     "    model_path=\"./models/xlm-roberta-base/model/xlm-roberta-base.p\",\n",
     "    model_config_path=\"./models/xlm-roberta-base/model/xlm-roberta-base.json\",\n",
-    "    model_tokenizer_path=\"./models/xlm-roberta-base/tokenizer\",\n",
     "    learning_rate=1e-5,\n",
     "    eval_every_steps=500,\n",
     "    do_train=True,\n",

diff --git a/guides/tutorials/quick_start_main.md b/guides/tutorials/quick_start_main.md
@@ -26,17 +26,15 @@ python jiant/scripts/download_data/runscript.py \
 2. Next, we download our RoBERTa-base model
 ```bash
 python jiant/proj/main/export_model.py \
-    --model_type ${MODEL_TYPE} \
+    --hf_pretrained_model_name_or_path ${MODEL_TYPE} \
     --output_base_path ${EXP_DIR}/models/${MODEL_TYPE}
 ```
 
 3. Next, we tokenize and cache the inputs and labels for our RTE task
 ```bash
 python jiant/proj/main/tokenize_and_cache.py \
     --task_config_path ${EXP_DIR}/tasks/configs/${TASK}_config.json \
-    --model_type ${MODEL_TYPE} \
-    --model_tokenizer_path \
-    ${EXP_DIR}/models/${MODEL_TYPE}/tokenizer \
+    --hf_pretrained_model_name_or_path ${MODEL_TYPE} \
     --output_dir ${EXP_DIR}/cache/${MODEL_TYPE}/${TASK} \
     --phases train,val \
     --max_seq_length 256 \