Unable to export GLM models to ONNX #35021
Comments
|
This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread. Please note that issues that do not follow the contributing guidelines are likely to be ignored. |
|
cc @Cyrilvallez |
|
I did a long deep-dive on this recently, and in fact, it appears to have been a bug with pytorch! pytorch/pytorch#145100 👀 I was able to apply a fix in Optimum by overriding the |
|
Hey @xenova! Thanks for looking deep into this. To be fair, |
Possibly... 👀 but to be sure, if you could post code snippets, I can test them out with the torch ONNX exporter. |
|
Sure, here are the class and functions with the modification: class GlmRotaryEmbedding(nn.Module):
def __init__(self, config: CohereConfig, device=None):
super().__init__()
# BC: "rope_type" was originally "type"
if hasattr(config, "rope_scaling") and config.rope_scaling is not None:
self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type"))
else:
self.rope_type = "default"
self.max_seq_len_cached = config.max_position_embeddings
self.original_max_seq_len = config.max_position_embeddings
self.config = config
self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device)
self.register_buffer("inv_freq", inv_freq, persistent=False)
self.original_inv_freq = self.inv_freq
def _dynamic_frequency_update(self, position_ids, device):
"""
dynamic RoPE layers should recompute `inv_freq` in the following situations:
1 - growing beyond the cached sequence length (allow scaling)
2 - the current sequence length is in the original scale (avoid losing precision with small sequences)
"""
seq_len = torch.max(position_ids) + 1
if seq_len > self.max_seq_len_cached: # growth
inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device, seq_len=seq_len)
self.register_buffer("inv_freq", inv_freq, persistent=False) # TODO joao: may break with compilation
self.max_seq_len_cached = seq_len
if seq_len < self.original_max_seq_len and self.max_seq_len_cached > self.original_max_seq_len: # reset
# This .to() is needed if the model has been moved to a device after being initialized (because
# the buffer is automatically moved, but not the original copy)
self.original_inv_freq = self.original_inv_freq.to(device)
self.register_buffer("inv_freq", self.original_inv_freq, persistent=False)
self.max_seq_len_cached = self.original_max_seq_len
@torch.no_grad()
def forward(self, x, position_ids):
if "dynamic" in self.rope_type:
self._dynamic_frequency_update(position_ids, device=x.device)
# Core RoPE block
inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1)
position_ids_expanded = position_ids[:, None, :].float()
# Force float32 (see https://github.com/huggingface/transformers/pull/29285)
device_type = x.device.type
device_type = device_type if isinstance(device_type, str) and device_type != "mps" else "cpu"
with torch.autocast(device_type=device_type, enabled=False):
freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
emb = torch.repeat_interleave(freqs, 2, dim=-1) # diff from Llama: we interleave() instead of cat()
cos = emb.cos()
sin = emb.sin()
# Advanced RoPE types (e.g. yarn) apply a post-processing scaling factor, equivalent to scaling attention
cos = cos * self.attention_scaling
sin = sin * self.attention_scaling
return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
def rotate_half(x):
"""Rotates half the hidden dims of the input."""
x1 = x[..., 0::2]
x2 = x[..., 1::2]
return torch.stack((-x2, x1), dim=-1).flatten(-2)
def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
"""Applies Rotary Position Embedding to the query and key tensors.
Args:
q (`torch.Tensor`): The query tensor.
k (`torch.Tensor`): The key tensor.
cos (`torch.Tensor`): The cosine part of the rotary embedding.
sin (`torch.Tensor`): The sine part of the rotary embedding.
position_ids (`torch.Tensor`, *optional*):
Deprecated and unused.
unsqueeze_dim (`int`, *optional*, defaults to 1):
The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
Returns:
`tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
"""
cos = cos.unsqueeze(unsqueeze_dim)
sin = sin.unsqueeze(unsqueeze_dim)
# Keep half or full tensor for later concatenation
rotary_dim = cos.shape[-1]
q_rot, q_pass = q[..., :rotary_dim], q[..., rotary_dim:]
k_rot, k_pass = k[..., :rotary_dim], k[..., rotary_dim:]
# Apply rotary embeddings on the first half or full tensor
q_embed = (q_rot * cos) + (rotate_half(q_rot) * sin)
k_embed = (k_rot * cos) + (rotate_half(k_rot) * sin)
# Concatenate back to full shape
q_embed = torch.cat([q_embed, q_pass], dim=-1)
k_embed = torch.cat([k_embed, k_pass], dim=-1)
return q_embed, k_embed |
System Info
transformersversion: 4.47.0.dev0Who can help?
@ArthurZucker @Cyrilvallez
Information
Tasks
examplesfolder (such as GLUE/SQuAD, ...)Reproduction
Run the following script:
It produces this error:
Expected behavior
The model should export correctly. This may in fact be an ONNX bug, but not 100% sure. Models like Gemma can export correctly, so it seems to be GLM-specific.
The text was updated successfully, but these errors were encountered: