SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model
Paper
•
2502.02737
•
Published
•
60
Nt3awnou
community
AI & ML interests
None defined yet.
Recent Activity
View all activity
nt3awnou's activity
loubnabnl
authored
a
paper
about 1 hour ago
alielfilali01
posted
an
update
about 1 month ago
Post
1923
3C3H AraGen Leaderboard welcomes today
deepseek-ai/DeepSeek-V3 and 12 other models (including the late gpt-3.5 💀) to the ranking of best LLMs in Arabic !
Observations:
- DeepSeek-v3 ranked 3rd and only Open model among the top 5 !
- A 14B open model ( Qwen/Qwen2.5-14B-Instruct) outperforms gpt-3.5-turbo-0125 (from last year). This shows how much we came in advancing and supporting Arabic presence within the LLM ecosystem !
- Contrary to what observed in likelihood-acc leaderboards (like OALL/Open-Arabic-LLM-Leaderboard) further finetuned models like maldv/Qwentile2.5-32B-Instruct actually decreased the performance compared to the original model Qwen/Qwen2.5-32B-Instruct.
It's worth to note that the decrease is statiscally insignificant which imply that at best, the out-domain finetuning do not really hurts the model original capabilities acquired during pretraining.
Previous work addressed this (finetuning VS pretraining) but more investigation in this regard is required (any PhDs here ? This could be your question ...)
Check out the latest rankings: inceptionai/AraGen-Leaderboard
Observations:
- DeepSeek-v3 ranked 3rd and only Open model among the top 5 !
- A 14B open model ( Qwen/Qwen2.5-14B-Instruct) outperforms gpt-3.5-turbo-0125 (from last year). This shows how much we came in advancing and supporting Arabic presence within the LLM ecosystem !
- Contrary to what observed in likelihood-acc leaderboards (like OALL/Open-Arabic-LLM-Leaderboard) further finetuned models like maldv/Qwentile2.5-32B-Instruct actually decreased the performance compared to the original model Qwen/Qwen2.5-32B-Instruct.
It's worth to note that the decrease is statiscally insignificant which imply that at best, the out-domain finetuning do not really hurts the model original capabilities acquired during pretraining.
Previous work addressed this (finetuning VS pretraining) but more investigation in this regard is required (any PhDs here ? This could be your question ...)
Check out the latest rankings: inceptionai/AraGen-Leaderboard
alielfilali01
posted
an
update
about 1 month ago
Post
1943
~75% on the challenging GPQA with only 40M parameters 🔥🥳
GREAT ACHIEVEMENT ! Or is it ?
This new Work, "Data Laundering: Artificially Boosting Benchmark Results through Knowledge Distillation", take out the mystery about many models i personally suspected their results. Speacially on leaderboards other than the english one, Like the Open Arabic LLM Leaderbaord OALL/Open-Arabic-LLM-Leaderboard.
The authors of this work, first started by training a model on the GPQA data, which, unsurprisingly, led to the model achieving 100% performance.
Afterward, they trained what they referred to as a 'legitimate' model on legitimate data (MedMCQA). However, they introduced a distillation loss from the earlier, 'cheated' model.
What they discovered was fascinating: the knowledge of GPQA leaked through this distillation loss, even though the legitimate model was never explicitly trained on GPQA during this stage.
This raises important questions about the careful use of distillation in model training, especially when the training data is opaque. As they demonstrated, it’s apparently possible to (intentionally or unintentionally) leak test data through this method.
Find out more: Data Laundering: Artificially Boosting Benchmark Results through Knowledge Distillation (2412.15255)
GREAT ACHIEVEMENT ! Or is it ?
This new Work, "Data Laundering: Artificially Boosting Benchmark Results through Knowledge Distillation", take out the mystery about many models i personally suspected their results. Speacially on leaderboards other than the english one, Like the Open Arabic LLM Leaderbaord OALL/Open-Arabic-LLM-Leaderboard.
The authors of this work, first started by training a model on the GPQA data, which, unsurprisingly, led to the model achieving 100% performance.
Afterward, they trained what they referred to as a 'legitimate' model on legitimate data (MedMCQA). However, they introduced a distillation loss from the earlier, 'cheated' model.
What they discovered was fascinating: the knowledge of GPQA leaked through this distillation loss, even though the legitimate model was never explicitly trained on GPQA during this stage.
This raises important questions about the careful use of distillation in model training, especially when the training data is opaque. As they demonstrated, it’s apparently possible to (intentionally or unintentionally) leak test data through this method.
Find out more: Data Laundering: Artificially Boosting Benchmark Results through Knowledge Distillation (2412.15255)
- 1 reply
alielfilali01
posted
an
update
about 2 months ago
Post
3460
Unpopular opinion: Open Source takes courage to do !
Not everyone is brave enough to release what they have done (the way they've done it) to the wild to be judged !
It really requires a high level of "knowing wth are you doing" ! It's kind of a super power !
Cheers to the heroes here who see this!
Not everyone is brave enough to release what they have done (the way they've done it) to the wild to be judged !
It really requires a high level of "knowing wth are you doing" ! It's kind of a super power !
Cheers to the heroes here who see this!
-
- 5 replies
alielfilali01
posted
an
update
about 2 months ago
Post
1539
Apparently i forgot to put this here !
Well, this is a bit late but consider given our recent blog a read if you are interested in Evaluation.
You don't have to be into Arabic NLP in order to read it, the main contribution we are introducing is a new evaluation measure for NLG. We made the fisrt application of this measure on Arabic for now and we will be working with colleagues from the community to expand it to other languages.
Blog:
Rethinking LLM Evaluation with 3C3H: AraGen Benchmark and Leaderboard
https://huggingface.co/blog/leaderboard-3c3h-aragen
Space:
inceptionai/AraGen-Leaderboard
Give it a read and let me know your thoughts 🤗
Well, this is a bit late but consider given our recent blog a read if you are interested in Evaluation.
You don't have to be into Arabic NLP in order to read it, the main contribution we are introducing is a new evaluation measure for NLG. We made the fisrt application of this measure on Arabic for now and we will be working with colleagues from the community to expand it to other languages.
Blog:
Rethinking LLM Evaluation with 3C3H: AraGen Benchmark and Leaderboard
https://huggingface.co/blog/leaderboard-3c3h-aragen
Space:
inceptionai/AraGen-Leaderboard
Give it a read and let me know your thoughts 🤗
Post
2231
Making SmolLM2 reproducible: open-sourcing our training & evaluation toolkit 🛠️ https://github.com/huggingface/smollm/
- Pre-training code with nanotron
- Evaluation suite with lighteval
- Synthetic data generation using distilabel (powers our new SFT dataset HuggingFaceTB/smoltalk)
- Post-training scripts with TRL & the alignment handbook
- On-device tools with llama.cpp for summarization, rewriting & agents
Apache 2.0 licensed. V2 pre-training data mix coming soon!
Which other tools should we add next?
- Pre-training code with nanotron
- Evaluation suite with lighteval
- Synthetic data generation using distilabel (powers our new SFT dataset HuggingFaceTB/smoltalk)
- Post-training scripts with TRL & the alignment handbook
- On-device tools with llama.cpp for summarization, rewriting & agents
Apache 2.0 licensed. V2 pre-training data mix coming soon!
Which other tools should we add next?
alielfilali01
posted
an
update
3 months ago
Post
2208
Unpopular opinion : o1-preview is more stupid than 4o and Qwen2.5-72B-Instruct in extremely underrated !
- 2 replies
alielfilali01
posted
an
update
4 months ago
Post
1715
I feel like this incredible resource hasn't gotten the attention it deserves in the community!
@clefourrier and generally the HuggingFace evaluation team put together a fantastic guidebook covering a lot about 𝗘𝗩𝗔𝗟𝗨𝗔𝗧𝗜𝗢𝗡 from basics to advanced tips.
link : https://github.com/huggingface/evaluation-guidebook
I haven’t finished it yet, but i'am enjoying every piece of it so far. Huge thanks @clefourrier and the team for this invaluable resource !
@clefourrier and generally the HuggingFace evaluation team put together a fantastic guidebook covering a lot about 𝗘𝗩𝗔𝗟𝗨𝗔𝗧𝗜𝗢𝗡 from basics to advanced tips.
link : https://github.com/huggingface/evaluation-guidebook
I haven’t finished it yet, but i'am enjoying every piece of it so far. Huge thanks @clefourrier and the team for this invaluable resource !
-
-
- 3 replies
alielfilali01
posted
an
update
4 months ago
Post
1842
Why nobdoy is talking about the new training corpus released by MBZUAI today.
TxT360 is +15 Trillion tokens corpus outperforming FineWeb on several metrics. Ablation studies were done up to 1T tokens.
Read blog here : LLM360/TxT360
Dataset : LLM360/TxT360
TxT360 is +15 Trillion tokens corpus outperforming FineWeb on several metrics. Ablation studies were done up to 1T tokens.
Read blog here : LLM360/TxT360
Dataset : LLM360/TxT360
-
- 2 replies
alielfilali01
posted
an
update
4 months ago
Post
2578
Don't you think we should add a tag "Evaluation" for datasets that are meant to be benchmarks and not for training ?
At least, when someone is collecting a group of datasets from an organization or let's say the whole hub can filter based on that tag and avoid somehow contaminating their "training" data.
At least, when someone is collecting a group of datasets from an organization or let's say the whole hub can filter based on that tag and avoid somehow contaminating their "training" data.
imomayiz
authored
a
paper
4 months ago
alielfilali01
posted
an
update
4 months ago
Post
892
We need a fork feature for models and datasets similar to "Duplicate this space" in spaces ! Don't you think ?
Sometimes you just want to save something in your profile privately and work on it later without the hassle of "load_.../push_to_hub" in a code file.
I know this is super lazy 😅 But it is what it is ...
tag : @victor
Sometimes you just want to save something in your profile privately and work on it later without the hassle of "load_.../push_to_hub" in a code file.
I know this is super lazy 😅 But it is what it is ...
tag : @victor
-
-
- 5 replies
alielfilali01
posted
an
update
4 months ago
Post
1212
@mariagrandury
(SomosNLP) and team releases the Spanish leaderboard !!!
It is impressive how they choosed to design this leaderboard and how it support 4 languages (all part of Spain ofc).
Check it out from this link :
la-leaderboard/la-leaderboard
It is impressive how they choosed to design this leaderboard and how it support 4 languages (all part of Spain ofc).
Check it out from this link :
la-leaderboard/la-leaderboard
- 1 reply
alielfilali01
posted
an
update
5 months ago
Post
381
This issue is just a treasure ! A bit deprecated i guess, but things are in their historical context. (personally, still need more to understand better)
https://github.com/huggingface/transformers/issues/8771
🫡 to the man @stas
https://github.com/huggingface/transformers/issues/8771
🫡 to the man @stas
alielfilali01
posted
an
update
5 months ago
Post
590
Are the servers down or what ? Am i the only one experiencing this error :
HfHubHTTPError: 500 Server Error: Internal Server Error for url: https://huggingface.co/api/datasets/...../)
Internal Error - We're working hard to fix this as soon as possible!-
-
- 2 replies
alielfilali01
posted
an
update
5 months ago
Post
1093
Datapluck: Portability Tool for Huggingface Datasets
"I found myself recently whipping up notebooks just to pull huggingface datasets locally, annotate or operate changes and update them again. This happened often enough that I made a cli tool out of it, which I've been using successfully for the last few months.
While huggingface uses open formats, I found the official toolchain relatively low-level and not adapted to quick operations such as what I am doing."
~ @omarkamali
Link : https://omarkama.li/blog/datapluck
"I found myself recently whipping up notebooks just to pull huggingface datasets locally, annotate or operate changes and update them again. This happened often enough that I made a cli tool out of it, which I've been using successfully for the last few months.
While huggingface uses open formats, I found the official toolchain relatively low-level and not adapted to quick operations such as what I am doing."
~ @omarkamali
Link : https://omarkama.li/blog/datapluck
- 1 reply
alielfilali01
posted
an
update
7 months ago
Post
723
Any idea if this "scheduled"/"dynamic" batch size is available in HF Trainers ? I've never seen it personally
loubnabnl
authored
a
paper
8 months ago
lekssays
authored
a
paper
8 months ago
Post
5869
🍷 FineWeb technical report is out and so is 📚 FineWeb-Edu, a 1.3 trillion tokens dataset that outperforms all other open web datasets, with remarkable improvements on educational benchmarks such as MMLU, ARC, and OpenBookQA.
Technical report: HuggingFaceFW/blogpost-fineweb-v1
Dataset: HuggingFaceFW/fineweb-edu
We used Llama 3 generations to train an educational quality classifier, filtering the 15 trillion tokens of FineWeb to select only those with high educational value (an approach also used in Llama 3 and Phi-3 training datasets). We're releasing both FineWeb-Edu and the classifier, along with a larger, less heavily filtered version containing 5.4 trillion tokens.
You can find more details about the dataset and the experiments we ran in the FineWeb technical report, It's a 45-minute read but it contains all the secret sauce for building high quality web datasets.
Enjoy!
Technical report: HuggingFaceFW/blogpost-fineweb-v1
Dataset: HuggingFaceFW/fineweb-edu
We used Llama 3 generations to train an educational quality classifier, filtering the 15 trillion tokens of FineWeb to select only those with high educational value (an approach also used in Llama 3 and Phi-3 training datasets). We're releasing both FineWeb-Edu and the classifier, along with a larger, less heavily filtered version containing 5.4 trillion tokens.
You can find more details about the dataset and the experiments we ran in the FineWeb technical report, It's a 45-minute read but it contains all the secret sauce for building high quality web datasets.
Enjoy!