Qwen3 sft collab #2355
Open
Qwen3 sft collab #2355
+290
−0
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
mydatascience
requested review from
gobbleturk,
khatwanimohit,
bvandermoon,
vipannalla,
RissyRan,
richjames0,
gagika,
shralex,
yangyuwei,
SurbhiJainUSC,
hengtaoguo,
A9isha,
aireenmei and
NuojCheng
as code owners
Signed-off-by: Vladimir Suvorov <[email protected]>
Signed-off-by: Vladimir Suvorov <[email protected]>
Signed-off-by: Vladimir Suvorov <[email protected]>
mydatascience
force-pushed
the
qwen3-sft-collab
branch
from
68582a2 to
e1cb7e4
Compare
Signed-off-by: Vladimir Suvorov <[email protected]>
Signed-off-by: Vladimir Suvorov <[email protected]>
Signed-off-by: Vladimir Suvorov <[email protected]>
Signed-off-by: Vladimir Suvorov <[email protected]>
| "print(f\"MaxText Home directory (from Python): {MAXTEXT_REPO_ROOT}\")\n", | ||
| "\n", | ||
| "DEBUG = False # set to True to run in debug mode, for more print statements\n", | ||
| "#set this to the path of the checkpoint you want to load, gs:// supported \n", |
There was a problem hiding this comment.
Line 118-121 is confusing. Let's simplify like this:
# Case 1: Set `MODEL_CHECKPOINT_PATH` to GCS path that already has `Qwen3-0.6B` model checkpoint
# Case 2: If you do not have the checkpoint, then do not update `MODEL_CHECKPOINT_PATH`
# and this colab will download the checkpoint from HF and store at `"{MAXTEXT_REPO_ROOT}/qwen_checkpoint\"`
"MODEL_CHECKPOINT_PATH = f\"{MAXTEXT_REPO_ROOT}/qwen_checkpoint\""
There was a problem hiding this comment.
Also, have you tried setting MODEL_CHECKPOINT_PATH to a GCS location? Do you see any permission issue to connect to GCS?
| "source": [ | ||
| "# This is the command to convert the HF model to the MaxText format \n", | ||
| "# You may omit it if you already have a checkpoint\n", | ||
| "!python3 -m MaxText.utils.ckpt_conversion.to_maxtext \\\n", |
There was a problem hiding this comment.
This command should only run when MODEL_CHECKPOINT_PATH = f\"{MAXTEXT_REPO_ROOT}/qwen_checkpoint. Put this behind a flag.
| " \"dtype=bfloat16\",\n", | ||
| " \"hf_path=HuggingFaceH4/ultrachat_200k\", # HuggingFace dataset/model if needed\n", | ||
| " f\"hf_access_token={HF_TOKEN}\",\n", | ||
| " \"base_output_directory=/tmp/maxtext_qwen06\",\n", |
There was a problem hiding this comment.
Can we set base_output_directory to a GCS path, so that users could access the fine-tuned checkpoint?
Also, add a print statement after train(), that you can find your fine-tuned checkpoint at this path.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
Qwen3 sft collab - an SFT collab with Qwen3-0.6B that can run on public collab 5e-1
Tests
Run the collab. Try yourself =)
Checklist
Before submitting this PR, please make sure (put X in square brackets):