Add a conversion script from huggingface to NeMo2 checkpoint format for

ESM-2.

Signed-off-by: Peter St. John <[email protected]>

source.txt

@@ -0,0 +1 @@
+('esm.embeddings.word_embeddings.weight', torch.float32, torch.Size([33, 320]))('esm.embeddings.position_embeddings.weight', torch.float32, torch.Size([1026, 320]))('esm.encoder.layer.0.attention.self.query.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.0.attention.self.query.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.0.attention.self.key.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.0.attention.self.key.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.0.attention.self.value.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.0.attention.self.value.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.0.attention.self.rotary_embeddings.inv_freq', torch.float32, torch.Size([8]))('esm.encoder.layer.0.attention.output.dense.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.0.attention.output.dense.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.0.attention.LayerNorm.weight', torch.float32, torch.Size([320]))('esm.encoder.layer.0.attention.LayerNorm.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.0.intermediate.dense.weight', torch.float32, torch.Size([1280, 320]))('esm.encoder.layer.0.intermediate.dense.bias', torch.float32, torch.Size([1280]))('esm.encoder.layer.0.output.dense.weight', torch.float32, torch.Size([320, 1280]))('esm.encoder.layer.0.output.dense.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.0.LayerNorm.weight', torch.float32, torch.Size([320]))('esm.encoder.layer.0.LayerNorm.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.1.attention.self.query.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.1.attention.self.query.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.1.attention.self.key.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.1.attention.self.key.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.1.attention.self.value.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.1.attention.self.value.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.1.attention.self.rotary_embeddings.inv_freq', torch.float32, torch.Size([8]))('esm.encoder.layer.1.attention.output.dense.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.1.attention.output.dense.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.1.attention.LayerNorm.weight', torch.float32, torch.Size([320]))('esm.encoder.layer.1.attention.LayerNorm.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.1.intermediate.dense.weight', torch.float32, torch.Size([1280, 320]))('esm.encoder.layer.1.intermediate.dense.bias', torch.float32, torch.Size([1280]))('esm.encoder.layer.1.output.dense.weight', torch.float32, torch.Size([320, 1280]))('esm.encoder.layer.1.output.dense.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.1.LayerNorm.weight', torch.float32, torch.Size([320]))('esm.encoder.layer.1.LayerNorm.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.2.attention.self.query.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.2.attention.self.query.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.2.attention.self.key.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.2.attention.self.key.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.2.attention.self.value.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.2.attention.self.value.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.2.attention.self.rotary_embeddings.inv_freq', torch.float32, torch.Size([8]))('esm.encoder.layer.2.attention.output.dense.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.2.attention.output.dense.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.2.attention.LayerNorm.weight', torch.float32, torch.Size([320]))('esm.encoder.layer.2.attention.LayerNorm.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.2.intermediate.dense.weight', torch.float32, torch.Size([1280, 320]))('esm.encoder.layer.2.intermediate.dense.bias', torch.float32, torch.Size([1280]))('esm.encoder.layer.2.output.dense.weight', torch.float32, torch.Size([320, 1280]))('esm.encoder.layer.2.output.dense.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.2.LayerNorm.weight', torch.float32, torch.Size([320]))('esm.encoder.layer.2.LayerNorm.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.3.attention.self.query.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.3.attention.self.query.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.3.attention.self.key.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.3.attention.self.key.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.3.attention.self.value.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.3.attention.self.value.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.3.attention.self.rotary_embeddings.inv_freq', torch.float32, torch.Size([8]))('esm.encoder.layer.3.attention.output.dense.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.3.attention.output.dense.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.3.attention.LayerNorm.weight', torch.float32, torch.Size([320]))('esm.encoder.layer.3.attention.LayerNorm.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.3.intermediate.dense.weight', torch.float32, torch.Size([1280, 320]))('esm.encoder.layer.3.intermediate.dense.bias', torch.float32, torch.Size([1280]))('esm.encoder.layer.3.output.dense.weight', torch.float32, torch.Size([320, 1280]))('esm.encoder.layer.3.output.dense.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.3.LayerNorm.weight', torch.float32, torch.Size([320]))('esm.encoder.layer.3.LayerNorm.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.4.attention.self.query.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.4.attention.self.query.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.4.attention.self.key.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.4.attention.self.key.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.4.attention.self.value.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.4.attention.self.value.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.4.attention.self.rotary_embeddings.inv_freq', torch.float32, torch.Size([8]))('esm.encoder.layer.4.attention.output.dense.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.4.attention.output.dense.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.4.attention.LayerNorm.weight', torch.float32, torch.Size([320]))('esm.encoder.layer.4.attention.LayerNorm.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.4.intermediate.dense.weight', torch.float32, torch.Size([1280, 320]))('esm.encoder.layer.4.intermediate.dense.bias', torch.float32, torch.Size([1280]))('esm.encoder.layer.4.output.dense.weight', torch.float32, torch.Size([320, 1280]))('esm.encoder.layer.4.output.dense.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.4.LayerNorm.weight', torch.float32, torch.Size([320]))('esm.encoder.layer.4.LayerNorm.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.5.attention.self.query.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.5.attention.self.query.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.5.attention.self.key.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.5.attention.self.key.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.5.attention.self.value.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.5.attention.self.value.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.5.attention.self.rotary_embeddings.inv_freq', torch.float32, torch.Size([8]))('esm.encoder.layer.5.attention.output.dense.weight', torch.float32, torch.Size([320, 320]))('esm.encoder.layer.5.attention.output.dense.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.5.attention.LayerNorm.weight', torch.float32, torch.Size([320]))('esm.encoder.layer.5.attention.LayerNorm.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.5.intermediate.dense.weight', torch.float32, torch.Size([1280, 320]))('esm.encoder.layer.5.intermediate.dense.bias', torch.float32, torch.Size([1280]))('esm.encoder.layer.5.output.dense.weight', torch.float32, torch.Size([320, 1280]))('esm.encoder.layer.5.output.dense.bias', torch.float32, torch.Size([320]))('esm.encoder.layer.5.LayerNorm.weight', torch.float32, torch.Size([320]))('esm.encoder.layer.5.LayerNorm.bias', torch.float32, torch.Size([320]))('esm.encoder.emb_layer_norm_after.weight', torch.float32, torch.Size([320]))('esm.encoder.emb_layer_norm_after.bias', torch.float32, torch.Size([320]))('esm.contact_head.regression.weight', torch.float32, torch.Size([1, 120]))('esm.contact_head.regression.bias', torch.float32, torch.Size([1]))('lm_head.bias', torch.float32, torch.Size([33]))('lm_head.dense.weight', torch.float32, torch.Size([320, 320]))('lm_head.dense.bias', torch.float32, torch.Size([320]))('lm_head.layer_norm.weight', torch.float32, torch.Size([320]))('lm_head.layer_norm.bias', torch.float32, torch.Size([320]))('lm_head.decoder.weight', torch.float32, torch.Size([33, 320]))

sub-packages/bionemo-esm2/src/bionemo/esm2/model/convert.py

@@ -0,0 +1,173 @@
+# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: LicenseRef-Apache2
+#
+# 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.
+
+
+from pathlib import Path
+
+import torch
+from nemo.lightning import io, teardown
+from nemo.lightning.pytorch.utils import dtype_from_hf
+from transformers import AutoConfig as HFAutoConfig
+from transformers import AutoModelForMaskedLM
+
+from bionemo.esm2.data.tokenizer import BioNeMoESMTokenizer, get_tokenizer
+from bionemo.esm2.model.model import ESM2Config
+from bionemo.llm.lightning import BionemoLightningModule
+from bionemo.llm.model.biobert.lightning import biobert_lightning_module
+
+
+@io.model_importer(BionemoLightningModule, "hf")
+class HFESM2Importer(io.ModelConnector[AutoModelForMaskedLM, BionemoLightningModule]):
+    """Converts a Hugging Face ESM-2 model to a NeMo ESM-2 model."""
+
+    def init(self) -> BionemoLightningModule:
+        """Initialize the converted model."""
+        return biobert_lightning_module(self.config, tokenizer=self.tokenizer)
+
+    def apply(self, output_path: Path) -> Path:
+        """Applies the transformation.
+
+        Largely inspired by
+        https://docs.nvidia.com/nemo-framework/user-guide/latest/nemo-2.0/features/hf-integration.html
+        """
+        source = AutoModelForMaskedLM.from_pretrained(str(self), trust_remote_code=True, torch_dtype="auto")
+        target = self.init()
+        trainer = self.nemo_setup(target)
+        self.convert_state(source, target)
+        self.nemo_save(output_path, trainer)
+
+        print(f"Converted ESM-2 model to Nemo, model saved to {output_path}")
+
+        teardown(trainer, target)
+        del trainer, target
+
+        return output_path
+
+    def convert_state(self, source, target):
+        """Converting HF state dict to NeMo state dict."""
+        mapping = {
+            # "esm.encoder.layer.0.attention.self.rotary_embeddings.inv_freq": "rotary_pos_emb.inv_freq",
+            "esm.encoder.layer.*.attention.output.dense.weight": "encoder.layers.*.self_attention.linear_proj.weight",
+            "esm.encoder.layer.*.attention.output.dense.bias": "encoder.layers.*.self_attention.linear_proj.bias",
+            "esm.encoder.layer.*.attention.LayerNorm.weight": "encoder.layers.*.self_attention.linear_qkv.layer_norm_weight",
+            "esm.encoder.layer.*.attention.LayerNorm.bias": "encoder.layers.*.self_attention.linear_qkv.layer_norm_bias",
+            "esm.encoder.layer.*.intermediate.dense.weight": "encoder.layers.*.mlp.linear_fc1.weight",
+            "esm.encoder.layer.*.intermediate.dense.bias": "encoder.layers.*.mlp.linear_fc1.bias",
+            "esm.encoder.layer.*.output.dense.weight": "encoder.layers.*.mlp.linear_fc2.weight",
+            "esm.encoder.layer.*.output.dense.bias": "encoder.layers.*.mlp.linear_fc2.bias",
+            "esm.encoder.layer.*.LayerNorm.weight": "encoder.layers.*.mlp.linear_fc1.layer_norm_weight",
+            "esm.encoder.layer.*.LayerNorm.bias": "encoder.layers.*.mlp.linear_fc1.layer_norm_bias",
+            "esm.encoder.emb_layer_norm_after.weight": "encoder.final_layernorm.weight",
+            "esm.encoder.emb_layer_norm_after.bias": "encoder.final_layernorm.bias",
+            "lm_head.dense.weight": "lm_head.dense.weight",
+            "lm_head.dense.bias": "lm_head.dense.bias",
+            "lm_head.layer_norm.weight": "lm_head.layer_norm.weight",
+            "lm_head.layer_norm.bias": "lm_head.layer_norm.bias",
+        }
+
+        # lm_head.bias
+        return io.apply_transforms(
+            source,
+            target,
+            mapping=mapping,
+            transforms=[_pad_embeddings, _pad_bias, _import_qkv_weight, _import_qkv_bias],
+        )
+
+    @property
+    def tokenizer(self) -> BioNeMoESMTokenizer:
+        """We just have the one tokenizer for ESM-2."""
+        return get_tokenizer()
+
+    @property
+    def config(self) -> ESM2Config:
+        """Returns the transformed ESM-2 config given the model tag."""
+        source = HFAutoConfig.from_pretrained(str(self), trust_remote_code=True)
+        output = ESM2Config(
+            num_layers=source.num_hidden_layers,
+            hidden_size=source.hidden_size,
+            ffn_hidden_size=source.intermediate_size,
+            position_embedding_type="rope",
+            num_attention_heads=source.num_attention_heads,
+            seq_length=source.max_position_embeddings,
+            fp16=(dtype_from_hf(source) == torch.float16),
+            bf16=(dtype_from_hf(source) == torch.bfloat16),
+            params_dtype=dtype_from_hf(source),
+        )
+
+        return output
+
+
+@io.state_transform(
+    source_key="esm.embeddings.word_embeddings.weight",
+    target_key="embedding.word_embeddings.weight",
+)
+def _pad_embeddings(ctx: io.TransformCTX, source_embed):
+    """Pad the embedding layer to the new input dimension."""
+    nemo_embedding_dimension = ctx.target.config.make_vocab_size_divisible_by
+    hf_embedding_dimension = source_embed.size(0)
+    num_padding_rows = nemo_embedding_dimension - hf_embedding_dimension
+    padding_rows = torch.zeros(num_padding_rows, source_embed.size(1))
+    return torch.cat((source_embed, padding_rows), dim=0)
+
+
+@io.state_transform(
+    source_key="lm_head.bias",
+    target_key="output_layer.bias",
+)
+def _pad_bias(ctx: io.TransformCTX, source_bias):
+    """Pad the embedding layer to the new input dimension."""
+    nemo_embedding_dimension = ctx.target.config.make_vocab_size_divisible_by
+    hf_embedding_dimension = source_bias.size(0)
+    output_bias = torch.zeros(nemo_embedding_dimension, dtype=source_bias.dtype, device=source_bias.device)
+    output_bias[:hf_embedding_dimension] = source_bias
+    return output_bias
+
+
+@io.state_transform(
+    source_key=(
+        "esm.encoder.layer.*.attention.self.query.weight",
+        "esm.encoder.layer.*.attention.self.key.weight",
+        "esm.encoder.layer.*.attention.self.value.weight",
+    ),
+    target_key="encoder.layers.*.self_attention.linear_qkv.weight",
+)
+def _import_qkv_weight(ctx: io.TransformCTX, query, key, value):
+    """Pad the embedding layer to the new input dimension."""
+    concat_weights = torch.cat((query, key, value), dim=0)
+    input_shape = concat_weights.size()
+    np = ctx.target.config.num_attention_heads
+    concat_weights = concat_weights.view(3, np, -1, query.size()[-1])
+    concat_weights = concat_weights.transpose(0, 1).contiguous()
+    concat_weights = concat_weights.view(*input_shape)
+    return concat_weights
+
+
+@io.state_transform(
+    source_key=(
+        "esm.encoder.layer.*.attention.self.query.bias",
+        "esm.encoder.layer.*.attention.self.key.bias",
+        "esm.encoder.layer.*.attention.self.value.bias",
+    ),
+    target_key="encoder.layers.*.self_attention.linear_qkv.bias",
+)
+def _import_qkv_bias(ctx: io.TransformCTX, query, key, value):
+    """Pad the embedding layer to the new input dimension."""
+    concat_biases = torch.cat((query, key, value), dim=0)
+    input_shape = concat_biases.size()
+    np = ctx.target.config.num_attention_heads
+    concat_biases = concat_biases.view(3, np, -1)
+    concat_biases = concat_biases.transpose(0, 1).contiguous()
+    concat_biases = concat_biases.view(*input_shape)
+    return concat_biases