Skip to content

Conversation

@MaxHao56
Copy link
Collaborator

  1. Graph fusion layer with attention mask = None
  2. sweeper = basics
  3. dataset bool

Optional Changes on the side. YAML file in the configs for giga pretrain needs to have match to your directory in you local or cluster machine

Copy link
Owner

@liamhebert liamhebert left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Few minor comments on root_dir, which when resolved are good to submit :)

@MaxHao56 MaxHao56 requested a review from liamhebert November 11, 2025 05:25
@MaxHao56 MaxHao56 marked this pull request as draft November 13, 2025 06:45
Copy link
Owner

@liamhebert liamhebert left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Still need to review EmbeddingGenerator, more soon.

"""Flatten a discussion tree into tensors for model input."""
dut.compute_relative_distance(tree)

flat = {
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We are missing a few parameters here. Namely, is_root and distances. These should be already attributes of node (I think).

MDT-2/src/tasks/dataset.py

Lines 860 to 867 in 5f50b4a

if is_root:
parent_id = node["id"]
if node["id"] not in result["id"]:
node["images"] = node["images"][0] if node["images"] else None
result["images"].append(node["images"])
result["distances"].append(node["distances"])

"out_degree": out_degree,
"attn_bias": torch.zeros((n, n)),
"distance": torch.zeros((n, n, 2)),
"distance_index": torch.zeros((n, n), dtype=torch.int16),
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This doesn't feel right... can you double check whether there should be a value here?

"input_ids": tokenized_text["input_ids"],
"attention_mask": tokenized_text["attention_mask"],
"token_type_ids": tokenized_text.get(
"token_type_ids", torch.zeros_like(tokenized_text["input_ids"])
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

likewise, we should be careful here

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants