Software Forgets: Agent Traces Are the Memory
• 6
AI & ML interests
The AI community building the future.
Recent Activity
View all activity
Papers
Qualixar OS: A Universal Operating System for AI Agent Orchestration
FineVision: Open Data Is All You Need
Articles
sayakpaul
updated a
dataset about 4 hours ago
enzostvs
updated a
dataset about 6 hours ago
alvarobartt
updated a
dataset about 6 hours ago
evalstate
updated a
bucket about 16 hours ago
Model search result sort is broken - Stuck on "trending"
1
#43 opened 3 days ago
by 
jijov17984
nielsr
updated a
Space about 16 hours ago
nielsr
updated a
Space about 17 hours ago
lysandre
updated a
dataset about 19 hours ago
Post
69
Two large changes in kernel-builder this week:
kernel-builder now links libstdc++ dynamically. To support a wide range of systems, we build against libstdc++ from manylinux_2_28 (EL 8 and later).
Following our Torch support policy that the current and previous Torch versions are supported, Torch 2.10 support was removed. We will soon also support the Torch stable ABI, so that it is possible to write kernels that support a large number of Torch versions.
kernel-builder now links libstdc++ dynamically. To support a wide range of systems, we build against libstdc++ from manylinux_2_28 (EL 8 and later).
Following our Torch support policy that the current and previous Torch versions are supported, Torch 2.10 support was removed. We will soon also support the Torch stable ABI, so that it is possible to write kernels that support a large number of Torch versions.
mishig
updated a
dataset about 21 hours ago
dlouapre
updated a
Space 1 day ago
Post
241
🤗 Announcing the Ettin Reranker family: six new state-of-the-art CrossEncoder rerankers for search from 17M to 1B parameters, plus the full training data and the ~150-line recipe. Built on the Ettin ModernBERT encoders, Apache 2.0. Details:
All six were trained with the same single-stage pointwise MSE distillation recipe, with mixedbread-ai/mxbai-rerank-large-v2 (1.54B) as the teacher. Only the learning rate and per-device batch size change between sizes. The 1B student matches the teacher within 0.0001 NDCG@10 on MTEB(eng, v2) Retrieval, the 150M is the strongest reranker I tested in the under-600M range, and the 17M beats the 33M ms-marco-MiniLM-L12-v2 by +0.051 NDCG@10 at roughly half the parameter count.
Speed matters as much as quality for a reranker, since it determines whether the model fits the latency budget between retrieval and showing results. Our 17M is the fastest reranker in the whole comparison at 7517 pairs/sec on an H100. Our 150M runs 2.3x faster than the two other 150M ModernBERT-base rerankers (gte-reranker-modernbert-base and granite-embedding-reranker-english-r2) because the modular Transformer module propagates unpadded inputs through every layer rather than just the FA2 attention kernel. And our 1B is 2.4x faster than its 1.5B teacher while matching it on quality.
I bootstrapped the training recipe with the new train-sentence-transformers Agent Skill shipped in Sentence Transformers v5.5.0. Install it with
I wrote a blog post walking through usage, results across six embedder pairings, the speed story, and the complete training script. Check it out, or just point your Agent to the URL:
https://huggingface.co/blog/ettin-reranker
Collection: https://huggingface.co/collections/cross-encoder/ettin-rerankers
All six were trained with the same single-stage pointwise MSE distillation recipe, with mixedbread-ai/mxbai-rerank-large-v2 (1.54B) as the teacher. Only the learning rate and per-device batch size change between sizes. The 1B student matches the teacher within 0.0001 NDCG@10 on MTEB(eng, v2) Retrieval, the 150M is the strongest reranker I tested in the under-600M range, and the 17M beats the 33M ms-marco-MiniLM-L12-v2 by +0.051 NDCG@10 at roughly half the parameter count.
Speed matters as much as quality for a reranker, since it determines whether the model fits the latency budget between retrieval and showing results. Our 17M is the fastest reranker in the whole comparison at 7517 pairs/sec on an H100. Our 150M runs 2.3x faster than the two other 150M ModernBERT-base rerankers (gte-reranker-modernbert-base and granite-embedding-reranker-english-r2) because the modular Transformer module propagates unpadded inputs through every layer rather than just the FA2 attention kernel. And our 1B is 2.4x faster than its 1.5B teacher while matching it on quality.
I bootstrapped the training recipe with the new train-sentence-transformers Agent Skill shipped in Sentence Transformers v5.5.0. Install it with
hf skills add train-sentence-transformers --claude and ask Claude Code (or Codex / Cursor / Gemini CLI) to fine-tune a SentenceTransformer, CrossEncoder, or SparseEncoder model on your data.I wrote a blog post walking through usage, results across six embedder pairings, the speed story, and the complete training script. Check it out, or just point your Agent to the URL:
https://huggingface.co/blog/ettin-reranker
Collection: https://huggingface.co/collections/cross-encoder/ettin-rerankers
alvarobartt
posted an update 3 days ago
Post
3125
Latest
TL; DR MoEs can be misleading to reason about from active parameters alone, since each token only activates a subset of experts, while the serving setup still needs to account for the full resident memory footprint.
🧠
🏗️ Dense models usually load and use most weights every forward pass, while MoEs load many experts but only route each token to a few of them
⚡ Active params isn't the same as memory footprint, especially for sparse architectures
📦 Runtime memory is about what is used per request/token, while loading memory also includes the expert weights that need to be resident
📚 KV cache can still dominate depending on context length, batch size, and concurrency
🔀 Expert Parallelism (EP) helps shard experts across accelerators when expert weights dominate
🚀 Data Parallelism (DP) + EP is often a good fit for throughput-oriented MoE serving
Check the repository at https://github.com/alvarobartt/hf-mem
hf-mem release added a breakdown of Mixture-of-Experts (MoE) memory usage!TL; DR MoEs can be misleading to reason about from active parameters alone, since each token only activates a subset of experts, while the serving setup still needs to account for the full resident memory footprint.
🧠
hf-mem now splits MoE memory into base model weights, routed experts, and KV cache🏗️ Dense models usually load and use most weights every forward pass, while MoEs load many experts but only route each token to a few of them
⚡ Active params isn't the same as memory footprint, especially for sparse architectures
📦 Runtime memory is about what is used per request/token, while loading memory also includes the expert weights that need to be resident
📚 KV cache can still dominate depending on context length, batch size, and concurrency
🔀 Expert Parallelism (EP) helps shard experts across accelerators when expert weights dominate
🚀 Data Parallelism (DP) + EP is often a good fit for throughput-oriented MoE serving
Check the repository at https://github.com/alvarobartt/hf-mem
Post
351
🤖 I've just published Sentence Transformers v5.5.0, headlined by a new train-sentence-transformers Agent Skill that lets your AI coding agent (Claude Code, Codex, Cursor, Gemini CLI, ...) train and finetune embedding, reranker, and sparse encoder models for you. Plus training losses & fixes. Details:
The skill bundles curated guidance for the whole training workflow across all three model types: base model selection, loss and evaluator choice, hard-negative mining, distillation, LoRA, Matryoshka, multilingual training, static embeddings, etc. It also ships production-ready training template scripts the agent can adapt. Install it with
On the loss side: EmbedDistillLoss is a new embedding-level distillation loss for SentenceTransformer. Instead of distilling teacher scores like MarginMSELoss, it aligns the student's embeddings directly with pre-computed teacher embeddings, wtih an optional learnable projection for when the student and teacher dimensions differ. Second, ADRMSELoss is a new listwise learning-to-rank loss for CrossEncoder from the Rank-DistilLLM paper, aimed at the LLM-distillation reranking setting.
encode() and predict() also gained a per-call processing_kwargs override, so you can change processor settings like max_length, a vision-language model's image resolution, or a video's fps, for a single call without rebuilding the model.
The Agent Skill is the part of this release I'm most keen for people to try. Curious to hear how it works for you. I've been using it myself a lot to quickly set up some training runs that immediately use a bunch of best practices.
> pip install sentence-transformers==5.5.0
> hf skills add train-sentence-transformers
The full release notes: https://github.com/huggingface/sentence-transformers/releases/tag/v5.5.0
The skill bundles curated guidance for the whole training workflow across all three model types: base model selection, loss and evaluator choice, hard-negative mining, distillation, LoRA, Matryoshka, multilingual training, static embeddings, etc. It also ships production-ready training template scripts the agent can adapt. Install it with
hf skills add train-sentence-transformers, then just describe what you want, e.g. "finetune a reranker on my (question, answer) pairs, mine hard negatives, and push it to the Hub".On the loss side: EmbedDistillLoss is a new embedding-level distillation loss for SentenceTransformer. Instead of distilling teacher scores like MarginMSELoss, it aligns the student's embeddings directly with pre-computed teacher embeddings, wtih an optional learnable projection for when the student and teacher dimensions differ. Second, ADRMSELoss is a new listwise learning-to-rank loss for CrossEncoder from the Rank-DistilLLM paper, aimed at the LLM-distillation reranking setting.
encode() and predict() also gained a per-call processing_kwargs override, so you can change processor settings like max_length, a vision-language model's image resolution, or a video's fps, for a single call without rebuilding the model.
The Agent Skill is the part of this release I'm most keen for people to try. Curious to hear how it works for you. I've been using it myself a lot to quickly set up some training runs that immediately use a bunch of best practices.
> pip install sentence-transformers==5.5.0
> hf skills add train-sentence-transformers
The full release notes: https://github.com/huggingface/sentence-transformers/releases/tag/v5.5.0
- 3 replies
sergiopaniego
posted an update 13 days ago
Post
1793
OpenEnv is growing fast in tutorials. If you're looking to get started with RL environments, check them out
> evaluate your agents using OpenEnv
> learn how rewards work via rubrics
> connect agents via MCP
> many moreeeee!
anything you think it's missing?
https://meta-pytorch.org/OpenEnv/tutorials/index.html
> evaluate your agents using OpenEnv
> learn how rewards work via rubrics
> connect agents via MCP
> many moreeeee!
anything you think it's missing?
https://meta-pytorch.org/OpenEnv/tutorials/index.html
sergiopaniego
posted an update 14 days ago
Post
815
OpenEnv already ships 🚢 with a ready-to-deploy RLM environment on free HF Spaces
Drop "Attention Is All You Need", write code that spawns parallel LLM calls → ✅ correct answer, reward 1.0, in 4.2s
Run GRPO (TRL) → model learns to write that search strategy itself
test it yourself → sergiopaniego/repl-env
check out OpenEnv → https://github.com/meta-pytorch/OpenEnv
Drop "Attention Is All You Need", write code that spawns parallel LLM calls → ✅ correct answer, reward 1.0, in 4.2s
Run GRPO (TRL) → model learns to write that search strategy itself
test it yourself → sergiopaniego/repl-env
check out OpenEnv → https://github.com/meta-pytorch/OpenEnv
AdinaY
submitted a
paper to Daily Papers about 1 month ago
sergiopaniego
posted an update about 1 month ago
Post
1384
Earlier this month, Apple introduced Simple Self-Distillation: a fine-tuning method that improves models on coding tasks just by sampling from the model and training on its own outputs with plain cross-entropy
And… it's already supported in TRL, built by Kashif Rasul. you can really feel the pace of development in the team 🐎
Paper by Ruixiang ZHANG, He Bai, Huangjie Zheng, Navdeep Jaitly, Ronan Collobert, Yizhe Zhang at Apple 🍎
How it works: the model generates completions at a training-time temperature (T_train) with top_k/top_p truncation, then fine-tunes on them with plain cross-entropy. no labels or verifier needed
You can try it right away with this ready-to-run example (Qwen3-4B on rStar-Coder):
https://github.com/huggingface/trl/blob/main/trl/experimental/ssd/ssd.py
or benchmark a checkpoint with the eval script:
https://github.com/huggingface/trl/blob/main/trl/experimental/ssd/ssd_eval.py
One neat insight from the paper: T_train and T_eval compose into an effective T_eff = T_train × T_eval, so a broad band of configs works well. even very noisy samples still help
Want to dig deeper?
Paper: Embarrassingly Simple Self-Distillation Improves Code Generation (2604.01193)
Trainer docs: https://huggingface.co/docs/trl/main/en/ssd_trainer
And… it's already supported in TRL, built by Kashif Rasul. you can really feel the pace of development in the team 🐎
Paper by Ruixiang ZHANG, He Bai, Huangjie Zheng, Navdeep Jaitly, Ronan Collobert, Yizhe Zhang at Apple 🍎
How it works: the model generates completions at a training-time temperature (T_train) with top_k/top_p truncation, then fine-tunes on them with plain cross-entropy. no labels or verifier needed
You can try it right away with this ready-to-run example (Qwen3-4B on rStar-Coder):
https://github.com/huggingface/trl/blob/main/trl/experimental/ssd/ssd.py
or benchmark a checkpoint with the eval script:
https://github.com/huggingface/trl/blob/main/trl/experimental/ssd/ssd_eval.py
One neat insight from the paper: T_train and T_eval compose into an effective T_eff = T_train × T_eval, so a broad band of configs works well. even very noisy samples still help
Want to dig deeper?
Paper: Embarrassingly Simple Self-Distillation Improves Code Generation (2604.01193)
Trainer docs: https://huggingface.co/docs/trl/main/en/ssd_trainer
sergiopaniego
posted an update about 1 month ago
Post
473
Great experience yesterday at PyTorch Conf Europe in Paris 🇫🇷
We (w/ @kashif ) talked about training LLMs through interaction, using trajectories across games, browsers, or simulators
Room was packed, a clear sign of interest in where RL post-training is heading.
sharing the slides! 🤓
https://drive.google.com/file/d/16k7YRnf5EJEo0XjXGlRJ_hVeLoFWKyNP/view?usp=sharing
We (w/ @kashif ) talked about training LLMs through interaction, using trajectories across games, browsers, or simulators
Room was packed, a clear sign of interest in where RL post-training is heading.
sharing the slides! 🤓
https://drive.google.com/file/d/16k7YRnf5EJEo0XjXGlRJ_hVeLoFWKyNP/view?usp=sharing
Post
1010
🌐 I've just published Sentence Transformers v5.4 to make the project fully multimodal for embeddings and reranking. The release also includes a modular CrossEncoder, and automatic Flash Attention 2 input flattening. Details:
You can now use SentenceTransformer and CrossEncoder with text, images, audio, and video, with the same familiar API. That means you can compute embeddings for an image and a text query using model.encode(), compare them with model.similarity(), and it just works. Models like Qwen3-VL-Embedding-2B and jinaai/jina-reranker-m0 are supported out of the box.
Beyond multimodal, I also fully modularized the CrossEncoder class. It's now a torch.nn.Sequential of composable modules, just like SentenceTransformer has been. This unlocked support for generative rerankers (CausalLM-based models like mxbai-rerank-v2 and the Qwen3 rerankers) via a new LogitScore module, which wasn't possible before without custom code.
Also, Flash Attention 2 now automatically skips padding for text-only inputs. If your batch has a mix of short and long texts, this gives you a nice speedup and lower VRAM usage for free.
I wrote a blog post walking through the multimodal features with practical examples. Check it out if you want to get started, or just point your Agent to the URL: https://huggingface.co/blog/multimodal-sentence-transformers
This release has set up the groundwork for more easily introducing late-interaction models (both text-only and multimodal) into Sentence Transformers in the next major release. I'm looking forward to it!
You can now use SentenceTransformer and CrossEncoder with text, images, audio, and video, with the same familiar API. That means you can compute embeddings for an image and a text query using model.encode(), compare them with model.similarity(), and it just works. Models like Qwen3-VL-Embedding-2B and jinaai/jina-reranker-m0 are supported out of the box.
Beyond multimodal, I also fully modularized the CrossEncoder class. It's now a torch.nn.Sequential of composable modules, just like SentenceTransformer has been. This unlocked support for generative rerankers (CausalLM-based models like mxbai-rerank-v2 and the Qwen3 rerankers) via a new LogitScore module, which wasn't possible before without custom code.
Also, Flash Attention 2 now automatically skips padding for text-only inputs. If your batch has a mix of short and long texts, this gives you a nice speedup and lower VRAM usage for free.
I wrote a blog post walking through the multimodal features with practical examples. Check it out if you want to get started, or just point your Agent to the URL: https://huggingface.co/blog/multimodal-sentence-transformers
This release has set up the groundwork for more easily introducing late-interaction models (both text-only and multimodal) into Sentence Transformers in the next major release. I'm looking forward to it!