Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fixes compatibility with Llama 3.3 70B #856

Open
wants to merge 5 commits into
base: main
Choose a base branch
from
Open

Fixes compatibility with Llama 3.3 70B #856

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
33ab01d
Add Mistral Small model and update Mistral rotary dimension
thisnick Jan 31, 2025
92e2224
Update pyproject.toml dependencies and configuration
thisnick Jan 31, 2025
505d1aa
Minor pyproject.toml formatting revert pyptojct.toml
thisnick Feb 5, 2025
89192e8
update model count
thisnick Feb 5, 2025
728c1b7
Add Mistral Small 24B model to Colab compatibility demo
thisnick Feb 5, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 14 additions & 13 deletions demos/Colab_Compatibility.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,9 +16,9 @@
"name": "stderr",
"output_type": "stream",
"text": [
"/var/folders/m3/z6c6rcdj1rbb2jh9vqpgvxg40000gn/T/ipykernel_57027/2944939757.py:18: DeprecationWarning: `magic(...)` is deprecated since IPython 0.13 (warning added in 8.1), use run_line_magic(magic_name, parameter_s).\n",
"/var/folders/pr/77j77_bs2gl2stxyrvr14x3c0000gn/T/ipykernel_60991/3507779555.py:18: DeprecationWarning: `magic(...)` is deprecated since IPython 0.13 (warning added in 8.1), use run_line_magic(magic_name, parameter_s).\n",
" ipython.magic(\"load_ext autoreload\")\n",
"/var/folders/m3/z6c6rcdj1rbb2jh9vqpgvxg40000gn/T/ipykernel_57027/2944939757.py:19: DeprecationWarning: `magic(...)` is deprecated since IPython 0.13 (warning added in 8.1), use run_line_magic(magic_name, parameter_s).\n",
"/var/folders/pr/77j77_bs2gl2stxyrvr14x3c0000gn/T/ipykernel_60991/3507779555.py:19: DeprecationWarning: `magic(...)` is deprecated since IPython 0.13 (warning added in 8.1), use run_line_magic(magic_name, parameter_s).\n",
" ipython.magic(\"autoreload 2\")\n"
]
}
Expand All @@ -43,7 +43,7 @@
" # Code to automatically update the HookedTransformer code as its edited without restarting the kernel\n",
" ipython.magic(\"load_ext autoreload\")\n",
" ipython.magic(\"autoreload 2\")\n",
" \n",
"\n",
"\n",
"\n",
"if IN_COLAB or IN_GITHUB:\n",
Expand All @@ -58,14 +58,14 @@
},
{
"cell_type": "code",
"execution_count": 1,
"execution_count": 2,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"TransformerLens currently supports 207 models out of the box.\n"
"TransformerLens currently supports 208 models out of the box.\n"
]
}
],
Expand Down Expand Up @@ -96,7 +96,7 @@
"def mark_models_as_tested(model_set: List[str]) -> None:\n",
" for model in model_set:\n",
" untested_models.remove(model)\n",
" \n",
"\n",
"\n",
"def run_set(model_set: List[str], device=\"cuda\") -> None:\n",
" for model in model_set:\n",
Expand All @@ -112,12 +112,12 @@
"def run_llama_set(model_set: List[str], weight_root: str, device=\"cuda\") -> None:\n",
" for model in model_set:\n",
" print(\"Testing \" + model)\n",
" # to run this, make sure weight root is the root that contains all models with the \n",
" # to run this, make sure weight root is the root that contains all models with the\n",
" # sub directories sharing the same name as the model in the list of models\n",
" tokenizer = LlamaTokenizer.from_pretrained(weight_root + model)\n",
" hf_model = LlamaForCausalLM.from_pretrained(weight_root + model, low_cpu_mem_usage=True)\n",
" tl_model = HookedTransformer.from_pretrained_no_processing(\n",
" model, \n",
" model,\n",
" hf_model=hf_model,\n",
" device=device,\n",
" fold_ln=False,\n",
Expand Down Expand Up @@ -309,7 +309,7 @@
"\n",
"if IN_COLAB:\n",
" run_set(free_compatible)\n",
" \n",
"\n",
"mark_models_as_tested(free_compatible)"
]
},
Expand Down Expand Up @@ -357,6 +357,7 @@
" \"mistralai/Mistral-7B-Instruct-v0.1\",\n",
" \"mistralai/Mistral-7B-v0.1\",\n",
" \"mistralai/Mistral-Nemo-Base-2407\",\n",
" \"mistralai/Mistral-Small-24B-Base-2501\",\n",
" \"Qwen/Qwen-7B\",\n",
" \"Qwen/Qwen-7B-Chat\",\n",
" \"Qwen/Qwen1.5-4B\",\n",
Expand All @@ -377,7 +378,7 @@
"\n",
"if IN_COLAB:\n",
" run_set(paid_gpu_models)\n",
" \n",
"\n",
"mark_models_as_tested(paid_gpu_models)"
]
},
Expand Down Expand Up @@ -410,7 +411,7 @@
"\n",
"if IN_COLAB:\n",
" run_set(paid_cpu_models, \"cpu\")\n",
" \n",
"\n",
"mark_models_as_tested(paid_cpu_models)"
]
},
Expand Down Expand Up @@ -528,7 +529,7 @@
"# Any models listed in the cell below have not been tested. This should always remain blank. If your\n",
"# PR fails due to this notebook, most likely you need to check any new model changes to ensure that\n",
"# this notebook is up to date.\n",
"print(*untested_models, sep = '\\n')"
"print(*untested_models, sep=\"\\n\")"
]
}
],
Expand All @@ -548,7 +549,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.9"
"version": "3.12.3"
}
},
"nbformat": 4,
Expand Down
3 changes: 2 additions & 1 deletion transformer_lens/loading_from_pretrained.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -182,6 +182,7 @@
"stabilityai/stablelm-tuned-alpha-7b",
"mistralai/Mistral-7B-v0.1",
"mistralai/Mistral-7B-Instruct-v0.1",
"mistralai/Mistral-Small-24B-Base-2501",
"mistralai/Mistral-Nemo-Base-2407",
"mistralai/Mixtral-8x7B-v0.1",
"mistralai/Mixtral-8x7B-Instruct-v0.1",
Expand Down Expand Up @@ -979,7 +980,7 @@ def convert_hf_model_config(model_name: str, **kwargs):
"normalization_type": "RMS",
"positional_embedding_type": "rotary",
"rotary_adjacent_pairs": False,
"rotary_dim": 32,
"rotary_dim": 128,
"final_rms": True,
"gated_mlp": True,
"rotary_base": 500000.0,
Expand Down
Loading