Daniel Johnson
We lifted 23 functions out of this paper's own repositories and ran 14 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| google-deepmind/treescope | canonical | 10 of 17 |
| google-deepmind/penzai | canonical | 4 of 6 |
| Function | Status | Where it lives |
|---|---|---|
| color_from_string | Ran | google-deepmind/treescope/treescope/formatting_util.py code served (permissive licence) · get_code("1ed46793bced4f6f") |
| dataclass_from_attributes | Ran | google-deepmind/treescope/treescope/dataclass_util.py code served (permissive licence) · get_code("491da2a807a0d703") |
| get_dtype_name | Ran | google-deepmind/treescope/treescope/dtype_util.py code served (permissive licence) · get_code("0bcd3e980f7fe655") |
| init_takes_fields | Ran | google-deepmind/treescope/treescope/dataclass_util.py code served (permissive licence) · get_code("3af90f8f7873fac9") |
| is_floating_dtype | Ran | google-deepmind/treescope/treescope/dtype_util.py code served (permissive licence) · get_code("07b1fce0fdc1eefd") |
| is_integer_dtype | Ran | google-deepmind/treescope/treescope/dtype_util.py code served (permissive licence) · get_code("5c689b98d9651bb5") |
| is_pytree_dataclass_type | Ran | google-deepmind/penzai/penzai/core/struct.py code served (permissive licence) · get_code("8da90bf2b11facb7") |
| oklch_color | Ran | google-deepmind/treescope/treescope/formatting_util.py code served (permissive licence) · get_code("49862fb509edca46") |
| parse_simple_color_and_pattern_spec | Ran | google-deepmind/treescope/treescope/formatting_util.py code served (permissive licence) · get_code("c54a7b3f3d04bf76") |
| styled | Ran | google-deepmind/treescope/treescope/figures.py code served (permissive licence) · get_code("cc38f8434830dd01") |
| tree_flatten_exactly_one_level | Ran | google-deepmind/penzai/penzai/core/tree_util.py code served (permissive licence) · get_code("7f0011a936783767") |
| var | Ran | google-deepmind/penzai/penzai/core/shapecheck.py code served (permissive licence) · get_code("c716cd4fc3d93251") |
| vars_for_axes | Ran | google-deepmind/penzai/penzai/core/shapecheck.py code served (permissive licence) · get_code("ad130894fc372aec") |
| with_font_size | Ran | google-deepmind/treescope/treescope/figures.py code served (permissive licence) · get_code("314cb1c24954c222") |
| default_well_known_filter | Not yet run | google-deepmind/treescope/treescope/canonical_aliases.py code served (permissive licence) · get_code("4eb59933d52ad100") |
| indented | Not yet run | google-deepmind/treescope/treescope/figures.py code served (permissive licence) · get_code("e4fd281c5d149c61") |
| is_pytree_node_field | Not yet run | google-deepmind/penzai/penzai/core/struct.py code served (permissive licence) · get_code("ffd28ac2c65414f3") |
| lookup_alias | Not yet run | google-deepmind/treescope/treescope/canonical_aliases.py code served (permissive licence) · get_code("5c75d7c159ec68b6") |
| maybe_defer_rendering | Not yet run | google-deepmind/treescope/treescope/lowering.py code served (permissive licence) · get_code("fe56c9c538b60fab") |
| maybe_local_module_name | Not yet run | google-deepmind/treescope/treescope/canonical_aliases.py code served (permissive licence) · get_code("ad8ee3a7572721bb") |
| pretty_keystr | Not yet run | google-deepmind/penzai/penzai/core/tree_util.py code served (permissive licence) · get_code("c45870654f09629f") |
| render_to_html_as_root | Not yet run | google-deepmind/treescope/treescope/lowering.py code served (permissive licence) · get_code("4fe2c2ee820dc796") |
| render_to_text_as_root | Not yet run | google-deepmind/treescope/treescope/lowering.py code served (permissive licence) · get_code("2a11ffd4903b882a") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Much of today's machine learning research involves interpreting, modifying or visualizing models after they are trained. I present Penzai, a neural network library designed to simplify model manipulation by representing models as simple data structures, and Treescope, an interactive prettyprinter and array visualizer that can visualize both model inputs/outputs and the models themselves. Penzai models are built using declarative combinators that expose the model forward pass in the structure of the model object itself, and use named axes to ensure each operation is semantically meaningful. With Penzai's tree-editing selector system, users can both insert and replace model components, allowing them to intervene on intermediate values or make other edits to the model structure. Users can then get immediate feedback by visualizing the modified model with Treescope. I describe the motivation and main features of Penzai and Treescope, and discuss how treating the model as data enables a variety of analyses and interventions to be implemented as data-structure transformations, without requiring model designers to add explicit hooks.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2408.00211")
get_code_for_paper("2408.00211")
have("2408.00211")
Connect an agent — have() is free.