Jacky Kwok on X: "Introducing Contrastive Language Model (CLM): an ultra-fast System One Model trained with a contrastive learning objective that connects states and actions. CLM-8B is pre-trained on internet-scale data and delivers up to 9× faster inference than Jev ⚡ while achieving comparable p… / X
Jacky Kwok on X: "Introducing Contrastive Language Model (CLM): an ultra-fast System One Model trained with a contrastive learning objective that connects states and actions. CLM-8B is pre-trained on internet-scale data and delivers up to 9× faster inference than Jev ⚡ while achieving comparable performance across computer-use, gaming, and tool-calling tasks. With lightweight fine-tuning, CLM-8B sets a new SOTA on challenging agentic coding benchmarks, such as DeepSWE (81.6%) and Terminal-Bench 2.1 (87.6%). In contrast, Jev fails to serve as an effective verifier for these long-horizon tasks. We also build an efficient training and serving infra for CLMs by disaggregating states and actions, allowing their embeddings to be cached and reused independently. This substantially reduces inference latency in settings where the state evolves continuously while the action set remains fixed. Finally, we establish scaling laws for CLMs and show that the test contrastive loss decreases predictably as a power law in training compute, model size, and dataset size.
📄 Blog: https://t.co/zwi9JOHKGx
💻 Code: https://t.co/rsHRYCGR8I
🗣️ Discord: https://t.co/Uqtdefvo3J
🤗 Data & Models: https://t.co/wdSWGGO3hu More details on CLM’s architecture, data recipe, and scaling laws in the thread below 🧵"
Introducing Contrastive Language Model (CLM): an ultra-fast System One Model trained with a contrastive learning objective that connects states and actions. CLM-8B is pre-trained on internet-scale data and delivers up to 9× faster inference than Jev ⚡ while achieving comparable performance across computer-use, gaming, and tool-calling tasks. With lightweight fine-tuning, CLM-8B sets a new SOTA on challenging agentic coding benchmarks, such as DeepSWE (81.6%) and Terminal-Bench 2.1 (87.6%). In contrast, Jev fails to serve as an effective verifier for these long-horizon tasks. We also build an efficient training and serving infra for CLMs by disaggregating states and actions, allowing their embeddings to be cached and reused independently. This substantially reduces inference latency in settings where the state evolves continuously while the action set remains fixed. Finally, we establish scaling laws for CLMs and show that the test contrastive loss decreases predictably as a power law in training compute, model size, and dataset size.
📄 Blog: contrastive-lm.notion.site
💻 Code: github.com/Contrastive-LM…
🗣️ Discord: discord.gg/5dAQEDJBs 🤗 Data & Models: huggingface.co/Contrastive-LM More details on CLM’s architecture, data recipe, and scaling laws in the thread below 🧵
Introducing Contrastive Language Model (CLM): an ultra-fast System One Model trained with a contrastive learning objective that connects states and actions. CLM-8B is pre-trained on internet-scale data and delivers up to 9× faster inference than Jev ⚡ while achieving comparable performance across computer-use, gaming, and tool-calling tasks. With lightweight fine-tuning, CLM-8B sets a new SOTA on challenging agentic coding benchmarks, such as DeepSWE (81.6%) and Terminal-Bench 2.1 (87.6%). In contrast, Jev fails to serve as an effective verifier for these long-horizon tasks. We also build an efficient training and serving infra for CLMs by disaggregating states and actions, allowing their embeddings to be cached and reused independently. This substantially reduces inference latency in settings where the state evolves continuously while the action set remains fixed. Finally, we establish scaling laws for CLMs and show that the test contrastive loss decreases predictably as a power law in training compute, model size, and dataset size.
📄 Blog: contrastive-lm.notion.site
💻 Code: github.com/Contrastive-LM…
🗣️ Discord: discord.gg/5dAQEDJBs 🤗 Data & Models: huggingface.co/Contrastive-LM More details on CLM’s architecture, data recipe, and scaling laws in the thread below 🧵
🧵(1) Model Architecture CLM first trains a state encoder and an action encoder on a large-scale dataset with a contrastive objective, so that each state is pulled toward the ground-truth action and pushed away from all others. The two encoders then serve directly as a zero-shotShow more
🧵(2) Data Recipe We release CLM-8B, which is pre-trained on 60M Nemotron Q&A pairs, mid-trained on 30M synthetic hard negatives, and post-trained on 1M agentic trajectories.
🧵(3) Scaling Laws for Verification We find that the test InfoNCE loss scales as a power law with training compute, dataset size, projection-head size, and encoder size. These dimensions must be scaled jointly to achieve optimal performance. Notably, scaling the encoder sizeShow more
🧵(4) Latency vs. Jev and Constrained Decoding The key difference between CLM and Jev is that Jev only supports state caching, whereas CLM’s dual-encoder architecture allows state and action embeddings to be cached independently. This is particularly useful in applications suchShow more
🧵(5) Zero-Shot Evaluation Across computer-use, gaming, and tool-calling tasks, CLM-8B performs on par with Jev while running up to 9× faster. The speedups are most pronounced when the number of candidates is large (e.g., WikiRacing) or when actions can be reused frequentlyShow more
🧵(6) Agentic Benchmarks We find that Jev fails to serve as a verifier for long-horizon tasks, performing below the random-selection (Pass@1) baseline. In contrast, with lightweight fine-tuning, CLM achieves SOTA performance on challenging agentic benchmarks, including DeepSWEShow more
CLM-8B is part of our scaling ladder, where we train models across multiple scales to establish scaling laws and predict performance at larger scales. A multimodal CLM-35B is now in training with more data, compute, and parameters. Stay tuned for the release early next month 🚀
the disaggregation is the part builders should steal: a loop scores the same state against many candidate actions, so the state embedding caches perfectly and each candidate collapses to a cheap dot product. that's the economics that makes per-turn decisions affordable.