PyTorch engagement report
@PyTorch - 519K followers on X
Measured over 46 original posts from a 30-day window, last computed on October 19, 2026.
Engagement
A typical post picks up 35 interactions against 519K followers, an engagement rate of 0.007%. Measured over 46 original posts, its engagement rate beats 29% of 20,444 tracked accounts of a similar size, which puts it in the middle of its size range rather than at either end. Posts are seen about 7.3K times each, and 0.477% of those impressions turn into an interaction. That is about 1.41% of the follower count, which is the gap between an audience on paper and an audience in a timeline. Posting runs at about 1.8 posts a day over the last 30 days, though only 43% of days saw any activity at all. Most posts go out around 16:00 UTC, and Thursday is the busiest day of the week. Of the 46 posts sampled, 98% carry an image or video and 100% link out. The account's strongest tracked post pulled 346 interactions, about 9.9x its own typical post. Recurring topics include #pytorchcon, #agntcon, #ai.
Measured over 46 original posts from a 30-day window, last computed on October 19, 2026. Recurring tags: #pytorchcon, #agntcon, #ai.
Compared with accounts its own size
PyTorch's engagement rate beats 29% of the tracked X accounts closest to it in follower count (20,444 accounts, accounts of similar size (decile 10 of 10)). A percentile is spread evenly by construction, so 50 really is the middle of that group and 90 really is its top tenth.
On engagement per impression rather than per follower it beats 30% of the same group. When those two numbers disagree, the gap is about how far its posts travel rather than how people react to them.
Where this sits in the catalog
At 0.007%, PyTorch sits above the 10th percentile of the 205,152 accounts in this comparison. That places it in the bottom 25% band, which runs below 0.024%.
Show the percentile table
| Percentile | Engagement rate |
|---|---|
| 10th percentile | 0.004% |
| 25th percentile | 0.024% |
| 50th percentile | 0.143% |
| 75th percentile | 0.678% |
| 90th percentile | 2.55% |
| 99th percentile | 82.4% |
This ruler is the whole measured catalog, not a size-matched group: it shows where the raw rate falls across every account we can measure, all of which are large. For a like-for-like comparison, read the size-band percentile above instead. See how the bands are built
Posting timing
This account posts most often around 16:00 UTC, and Thursday is its busiest day of the week. The bars below are the catalog-wide pattern, with this account's own busiest slot marked. They do not show how this account performs at each hour: we keep one aggregate per account, not one per hour, so that measurement does not exist in our data.
Show engagement by hour posted, utc as a table
| Hour (UTC) | Vs author median | Posts |
|---|---|---|
| 00:00 UTC | -1% | 89K |
| 01:00 UTC | -2% | 90K |
| 02:00 UTC | -3% | 88K |
| 03:00 UTC | -4% | 94K |
| 04:00 UTC | -5% | 76K |
| 05:00 UTC | -4% | 75K |
| 06:00 UTC | -5% | 86K |
| 07:00 UTC | -5% | 93K |
| 08:00 UTC | -4% | 108K |
| 09:00 UTC | -4% | 124K |
| 10:00 UTC | -3% | 129K |
| 11:00 UTC | -3% | 141K |
| 12:00 UTC | -3% | 154K |
| 13:00 UTC | -3% | 167K |
| 14:00 UTC | -4% | 173K |
| 15:00 UTC | -2% | 176K |
| 16:00 UTC | -3% | 171K |
| 17:00 UTC | -3% | 159K |
| 18:00 UTC | -2% | 149K |
| 19:00 UTC | -2% | 141K |
| 20:00 UTC | -1% | 131K |
| 21:00 UTC | 0% | 116K |
| 22:00 UTC | -2% | 100K |
| 23:00 UTC | -1% | 90K |
Show engagement by day of week as a table
| Day | Vs author median | Posts |
|---|---|---|
| Sunday | +5% | 393K |
| Monday | +1% | 483K |
| Tuesday | -2% | 520K |
| Wednesday | -3% | 472K |
| Thursday | -2% | 430K |
| Friday | -3% | 447K |
| Saturday | +2% | 393K |
Formats this account uses
Its own posting mix on the left, and what each of those formats does across every account we track on the right. Only formats where the effect clears our publish test appear here, so an empty row is a format we could not measure rather than one that does nothing.
| Format | This account | Catalog effect | 95% interval | Accounts behind it |
|---|---|---|---|---|
| Image or video | 98% of posts | +111% | +108% to +115% | 34K |
| Outbound link | 100% of posts | -41% | -42% to -40% | 32K |
| Typical length | - | +15% | +14% to +16% | 32K |
- 98% of this account's sampled posts carry an image or video. Across the catalog, posts with an image or video run 111% above the same accounts' other posts.
- 100% of its posts carry a link off X. Across the catalog, posts with an outbound link run 41% below the same accounts' other posts, so a large share of this account's output sits in the weakest bucket we measure.
- Its average post runs 472 characters, which falls in the over 280 characters band. Across the catalog, posts over 280 characters run 15% above the same accounts' other posts.
These are catalog-wide differences applied to this account's own posting mix, not a measurement of how each format performs for this account specifically. We keep one median per account, not one per format per account, so the second thing is not something this data can tell you.
Best tweets
- Sep 3, 20269.9x their median
Building robotaxis or L4 autonomy? Don't miss our livestream. On Sept. 9 @ 9AM PT, we'll deep dive into NVIDIA Alpamayo 2 Super, covering release notebooks, robotaxi pipelines, AlpaSim updates and more. Save the date: https://t.co/Z9r0c3P1wb https://t.co/fv646RGPyB
- Aug 10, 20265.7x their median
Today @AIatMeta introduced Muse Glimmer, an open-weight, 30-billion-parameter model distilled from Meta’s Muse Spark for on-device agentic workflows. Alongside, ExecuTorch is adding end-to-end support for running Muse Glimmer on @NVIDIA GPUs and Macs with @Apple silicon. Why ExecuTorch? ExecuTorch offers a smarter, native approach: ⚡ Build in PyTorch: Implement your model and decoding strategies right in PyTorch as you normally would ⚡ Seamless Export: Export directly to ExecuTorch when you're ready for deployment ⚡ Automated Optimization: The framework automatically handles backend-specific lowering (Triton on CUDA, MLX-native, or Metal on Apple silicon) and uses ahead-of-time compilation to optimize the full execution path end-to-end Learn more: https://t.co/ABzlzreEJz
- Oct 16, 20265.4x their median
Zephon’s TorchTitan integration preserves identical global batches when training resumes on a different GPU count. In our latest blog’s 1B-parameter model test, training resumed from a 16-GPU checkpoint on 8, 32, and 64 GPUs, and the SHA-256 hash of every global batch matched the original run. Zephon tokenizes, packs, shuffles, and mixes data inside the training job, avoiding a preprocessed copy of the corpus. The data, tokenizer, seed, mixture, context length, and global batch size must stay unchanged when resuming. The lane count must divide evenly across data-parallel replicas, and each optimizer step’s batch must split evenly across lanes. Training losses can still differ because floating-point reduction order depends on GPU count. ✍️ @datologyai @josh_wills @MaxiBoether, Ties Robroek (@datologyai) Read more: https://t.co/JKn45hQIgU
- Sep 11, 20265.4x their median
Validate your real-world PyTorch skills with the PyTorch Certified Associate (PTCA) program. Designed for early-stage practitioners, PTCA helps you demonstrate foundational ability across model design, training, and deployment while building credibility across the open source AI ecosystem. Learn more and enroll here: https://t.co/JJe21Phgup
- Aug 30, 20265.3x their median
vLLM @vllm_project sessions at #PyTorchCon NA 2026 span the serving stack, from attention and KV cache management to disaggregated serving and hardware portability across accelerators. Speakers will cover attention and KV cache systems, disaggregated serving, expert parallelism, and vLLM across TPU, Trainium, Arm, and IBM Spyre. Featured speakers include: @RedHat: Lucas Wilkinson, Matthew Bonanni, Zhanqiu Hu, @rickynds, and Alex Brooks @amazon: Sunita Nadampalli @IBM: Or Ozeri, Thomas Parnell, and Dave Grove @Huawei: Mengqing Cao @nvidia: Itay Alroy @Google: @Rob_Mulla and Qi Zhou @awscloud: Maen Suleiman @Meta: Richard Zou, Colin Taylor, and Angela Yi @BAAIBeijing: Yonghua Lin @fujitsulabs: Abhishek Jain and N Maajid Khan @IBMResearch: Antoni Viros i Martin and Avery Blanchard @MistralAI: Nicolò Lucchesi @tensormesh: @this_will_echo @googlecloud: Bill Jia Register by September 4 to save: https://t.co/GIZUAha5wA Read the vLLM session guide: https://t.co/d2ozJTwa2G
- Sep 8, 20264.7x their median
Major ecosystem libraries like PyTorch and CuPy now build on the shared cuda.core foundation, reducing version conflicts and interop bugs for developers. With CUDA 13.3, @nvidia released CUDA Python 1.0, the libraries and tools that give you the full CUDA platform from Python. Using standard CUDA context, devices, streams, and memory are shared with the rest of the Python GPU ecosystem, so your own kernels can run against CuPy arrays or PyTorch tensors without copying data. Read the full blog post: https://t.co/UNoXigEQBd
- Oct 6, 20264.6x their median
Replacing CUDA with FBTriton for Table Batched Embedding (TBE) kernels delivered massive gains for our rec sys models. @Meta contributors achieved up to 1.28x faster forward passes, 2x faster backward passes, and a huge boost in developer velocity. Check out the engineering deep-dive: https://t.co/qdP0nnoLgF ✍️ Daohang Shi, Oleksandr Stashuk, Rupert Wu, Liangbei Xu, Rich Zhu
- Sep 2, 20264.4x their median
PyTorch 2.14 brings CuTeDSL-generated CUTLASS kernels to Inductor through NVGEMM, a new nccl2 backend for PyTorch Distributed, first-class c10d fault tolerance, native linear algebra on Apple Silicon, torch.switch and CUDA graph capture for torch.while_loop, declarative dynamic shapes via @ dynamic_spec, broader platform support, and experimental torch.compile support for complex-valued tensors. Highlights include: - NVGEMM brings CuTeDSL-generated CUTLASS kernels to Inductor, with epilogue fusion, scaled and NVFP4 GEMM, and grouped-reduction epilogues autotuned alongside Triton and ATen - A new nccl2 backend for PyTorch Distributed, ported from torchcomms, implementing the full collective contract with nonblocking communicators and eager communicator splitting - Fault tolerance becomes a first-class c10d concept, with in-place process-group reconfiguration, one-sided RMA windows, and a Flight Recorder that works for any backend rather than only NCCL - Apple Silicon gains native linear algebra, including Jacobi-kernel SVD, eigh, QR and Cholesky, alongside a five-part reduction rewrite and a further MPSGraph to Metal kernel migration - torch.switch generalizes torch.cond to multi-way branching, and torch.while_loop can now be captured in a CUDA graph - Declarative dynamic shapes via @ dynamic_spec, shared across torch.compile, torch.export and make_fx - Broader platform support: ROCm 7.14 wheels are produced from the TheRock pip SDK, Intel XPU adds native graph capture, and Inductor targets Rubin (sm_107) - Experimental torch.compile support for complex-valued tensors: opt-in support decomposes supported complex operations into real and imaginary computations, enabling compiler backends to optimize more complex-number workloads PyTorch 2.14 includes 2,995 commits from 487 contributors since PyTorch 2.13. Bring your questions about PyTorch 2.14 to our live Q&A on Thursday, September 17, 2026 at 11 AM PT. Andrey Talman (@Meta), Natalia Gimelshein (@Meta), Joe Spisak (@joespeez, @reflection_ai), and Chris Gottbrath (Gottbrath Tech) will share an overview of PyTorch 2.14 and answer community questions about PyTorch and the new capabilities. Register for the live Q&A: https://t.co/kDoOVMUCU0 🔗 Read the PyTorch 2.14 release blog: https://t.co/bcqC8jrIlW
- Oct 13, 20264.3x their median
What does it take to support new AI hardware within PyTorch? In our latest blog, contributors from @IBM share how they extended PyTorch Inductor for the IBM Spyre Accelerator, leveraging PyTorch's compiler to address challenges in data movement, memory management, and work distribution while optimizing for Spyre's unique architecture. Read on for a look at how PyTorch's extensibility helps developers bring support for new hardware to the framework: https://t.co/m28U0pTSr7
- Sep 17, 20264.2x their median
For supported biomolecular structure-prediction models, NVIDIA BioNeMo Inference Runtime keeps the familiar PyTorch workflow and lets developers construct supported models as torch.nn.Module objects or reuse selected modules in custom code. In a new NVIDIA Developer Blog post, the team details how BioIR uses optimized kernels and, where applicable, CUDA Graphs for model execution, with Ray replicas for processing large batches of independent inputs across GPUs in a single node. This post covers BioIR acceleration for Boltz-2, OpenFold2, and OpenFold3. Read more: https://t.co/KUr1DCIjXX
Ranked by total interactions across everything we have tracked for this account, which is a longer history than the 30-day window the rates above use. The multiple compares each post to this account's own median.
Recurring topics
The most frequent hashtags in the sampled posts. They describe what this account writes about; they are not a performance signal, and the catalog-wide breakdown on the hub shows how little hashtag count moves.
Buy or sell Twitter (X) accounts - escrow-protected
PlayerSells is an escrow marketplace for Twitter (X) accounts. Every deal is protected, with no middleman risk.
Reading these numbers
A typical post picks up 35 interactions against 519K followers, an engagement rate of 0.007%. Measured over 46 original posts, its engagement rate beats 29% of 20,444 tracked accounts of a similar size, which puts it in the middle of its size range rather than at either end. Posts are seen about 7.3K times each, and 0.477% of those impressions turn into an interaction. That is about 1.41% of the follower count, which is the gap between an audience on paper and an audience in a timeline. Posting runs at about 1.8 posts a day over the last 30 days, though only 43% of days saw any activity at all. Most posts go out around 16:00 UTC, and Thursday is the busiest day of the week. Of the 46 posts sampled, 98% carry an image or video and 100% link out. The account's strongest tracked post pulled 346 interactions, about 9.9x its own typical post. Recurring topics include #pytorchcon, #agntcon, #ai.
- What is PyTorch's engagement rate on X?
- PyTorch (@PyTorch) has an engagement rate of 0.007%, based on the median interactions across 46 original posts from the last 30 days against 518,597 followers. Replies, reposts and quote-posts of other people are excluded from that sample.
- Is that a good engagement rate?
- At 0.007%, PyTorch sits above the 10th percentile of the 205,152 accounts in this comparison. Those comparison accounts are all large ones, because our scanning cadence is weighted towards big accounts, so this is a ranking among peers of similar scale rather than a ranking across X.
- Does @PyTorch have real engagement?
- Its engagement rate beats 29% of the tracked X accounts closest to it in follower count (20,444 accounts), which puts it in the middle of its size range group. Ranking inside a size band matters because engagement rate falls as accounts grow, so a raw rate would mostly re-measure the follower count. It is a starting point for a look at follower quality, not a verdict on it.
- When does @PyTorch post?
- Most posts go out around 16:00 UTC, and Thursday is its busiest day, at roughly 1.83 posts per day across the measured window.