[ICLR'26] The official code implementation for "Cache-to-Cache: Direct Semantic Communication Between Large Language Models"
C-Rank Grade: D (Risky) - 10/100
External PR Merge Rate: -
Response Time: 1d
First Timer Success: -
thu-nics/C2C has a recorded first-timer success rate of 0.0%. Repositories ranked D typically provide actionable feedback during code reviews and actively nurture new community contributors.
Maintainers in thu-nics/C2C respond to incoming external pull requests in approximately 34.2 hours on average. Keeping PRs focused on single tasks and ensuring tests pass helps maintainers review faster.
The C-Rank™ system evaluates GitHub projects on a 0–100 scale using real data: PR merge rates, review turnaround time, active maintainer presence, and first-time contributor success. A score of 10.2 places thu-nics/C2C in the D tier.
The external contributor pull request merge rate for thu-nics/C2C is 0.0%, based on public PR activity from non-core contributors.
thu-nics/C2C currently has 2 active issue(s) tagged with beginner-friendly labels like "good first issue", "beginner", or "up-for-grabs".
Collaborative maintainer environment in thu-nics/C2C. Reviews community pull requests with focus on project quality.
Discussions in thu-nics/C2C focus heavily on practical implementation feedback, code formatting standards, and issue reproduction details.
Unlinked PRs without issue context and changes that fail automated test suites face the highest review friction.
Average Response Latency
Tracks hours until a maintainer leaves a review, comment, or PR response.
External Acceptance Rate
Percentage of community pull requests successfully merged into main.
First PR Conversion
Rate at which developers submitting their first repository PR succeed.
Show contributors that your repository actively reviews and merges external pull requests.
[](https://getmerged.abhishekco.de/thu-nics/C2C?utm_source=github&utm_medium=badge)
External PR acceptance
First PR success rate
50% receive response in
90% receive response in
Maintainer review activity heatmap
Maintainer Strategy: Maintainer Sweet Spot: PRs under 150 lines merge in ~22 hours (81% approval). Large PRs (>500 lines) take ~164 hours (29% approval). Keep your diff compact for rapid review.
Very cool work! I trained a C2C fuser with Qwen3-1.7B as receiver and Qwen3-8B as sharer, following the paper’s OpenHermes-500k configuration as closely as possible. Model MMLU-Redux OpenBookQA ARC-C C-Eval Receiver 58.27 65.20 75.30 54.62 Sharer 75.59 85.20 90.87 73.46 C2C 53.76 64.00 72.96 53.08 C2C consistently underperforms the receiver. Since the paper does not report this exact pairing, I may have missed a model-specific setting or evaluation detail. However, I was able to successfully retrain the pairings for which you published Huggingface checkpoints. Do you have any advice on hyperparameters that might improve training, or any detail that I may have missed? My wandb run
Hi @giedlinlofty118-blip, thanks for the detailed follow-up. I rechecked the experiment settings for the GSM8K Table 15 run, and I need to correct one important point in my previous reply. The main source of confusion is the fuser checkpoint. The 62.55 GSM8K C2C result in Table 15 was obtained with a GSM8K-specific C2C fuser trained for the Qwen3-0.6B receiver and Qwen3-4B sharer. It was not obtained with the public generic qwen3_0.6b+qwen3_4b_Fuser checkpoint. Exact Setup for the 62.55 C2C Row Receiver/base model: Qwen/Qwen3-0.6B Sharer model: Qwen/Qwen3-4B Fuser: a GSM8K-specific C2C fuser trained from Qwen3-4B generated GSM8K training data Evaluation dataset: full openai/gsm8k test split So the key correction is: Table 15 used a task-specific GSM8K fuser, not the public general-purpose Qwen3-0.6B + Qwen3-4B fuser. Below are the settings used for that run. 1. Teacher Data Generation for the GSM8K Fuser The GSM8K-specific fuser was trained on GSM8K train examples with responses gene
Rate what actually matters after opening a pull request here.
Rate your first-hand PR experience (review speed, maintainer responsiveness, and onboarding ease) to help other contributors.
When evaluating whether to contribute to thu-nics/C2C, response velocity and maintainer engagement are crucial. GetMerged continuously tracks pull request trajectories, first-comment latency, and code review rounds to help developers avoid submitting pull requests to backlogged repositories.
Currently, maintainers of thu-nics/C2C acknowledge new external contributions in approximately 1 days+. Out of all submitted pull requests from non-core authors in the last 180-day window, 0.0% were successfully merged into the primary branch.
thu-nics/C2C has a recorded first-timer success rate of 0.0%. Repositories ranked Risky typically provide actionable feedback during code reviews and actively nurture new community contributors.
The initial maintainer response time averages ~1 days+. Keeping PRs scoped to single concerns and ensuring CI checks succeed will optimize review turnaround.
The C-Rank™ index scores repositories on a 0 to 100 scale using an objective formula: external PR merge rates, initial response speed, active maintainer count, and first-time contributor retention. A score of 10.2 places thu-nics/C2C in the Risky tier.
The external pull request merge rate is 0.0%. GetMerged isolates non-core community contributions so external developers get an accurate benchmark of PR acceptance probability.
Yes, thu-nics/C2C currently has 2 active issue(s) tagged with beginner-friendly labels. You can inspect these directly from the repository issues tab.
All metrics displayed for thu-nics/C2C are automatically retrieved via the public GitHub API and recalculated daily. Insider pull requests submitted by repository owners or organization members are excluded from merge rate calculations to preserve objective external contributor statistics.