Enterprise AI + omnichannel communications. 25+ years. GSMA member. AI infra | research | agents, voice, vision https://xitter.cf/t.co/Rz5PUzbh4R
Global
Joined December 2015
- Tweets3.2K
- Following290
- Followers984
- Likes518
🎃 Five apps. Fifty messages. One terrifying inbox.
👻 Don't let customer conversations disappear into the dark.
One inbox. Less chaos. More conversations.
👉 broadnet.me
#BroadNet #Omnichannel #Halloween
Your AI shouldn’t just answer. It should get things done. 🚀
From updating records to scheduling meetings and automating workflows, BroadNet AI turns requests into real actions.
Less manual work. Faster execution. Smarter business operations.
⚡ Stop asking. Start getting things done.
Discover more → broadnet.ai
#BroadNetAI #AIAutomation #AIForBusiness #WorkflowAutomation #BusinessAI #DigitalTransformation
BroadNet retweeted
Supabase phone sign-up codes don't have to go out by SMS.
Connect MyOTP as your Send SMS Hook and pick WhatsApp or Telegram for the codes instead. Same hook, billed at your plan's rate for that channel. Three steps, no code.
myotp.app/integrations/supab…
🎬 Black Friday deserves a grand entrance!
Get ahead of the rush and make your offers stand out with BroadNet’s Business Communication Solutions.🚀
📩 Ready to launch your campaign?Contact us at sales@broadnet.me
#BlackFriday2026 #BroadNet #BusinessMessaging #BlackFridayMarketing
Your brand. Your clients. Your business. 🚀
We power the SMS platform. You make it your own!
Start reselling with BroadNet’s White-Label SMS Solution.
📩 sales@broadnet.me | broadnet.me
#BroadNet #WhiteLabelSMS #SMSReseller #BusinessMessaging
🎃 Too many channels? That's scary!
One inbox. Every conversation. Zero nightmares. 👻
Keep it all under control with BroadNet Omnichannel.
sales@broadnet.me
#BroadNet #Omnichannel #Halloween
🎃 Too many channels? That's scary!
One inbox. Every conversation. Zero nightmares. 👻
Keep it all under control with BroadNet Omnichannel.
broadnet.me
#BroadNet #Omnichannel #Halloween
Too many channels? One simple solution.
BroadNet Omnichannel brings customer conversations together in one place.
💬 Connect smarter.
🌐 broadnet.me
#BroadNet #Omnichannel
All routes lead to Muscat. The opportunities? Endless! 🌍
BroadNet is heading to Middle East 2026 GCCM & SIS Summit to connect, exchange ideas, and build stronger global partnerships.
📍 Muscat, Oman | November 1–4, 2026
🤝Let's turn one conversation into the next big opportunity!
📩 bit.ly/3WdABsk | events@broadnet.me
#BroadNet #GCCM2026 #GlobalConnectivity #TelecomPartnerships #BusinessMessaging
Long documents too: 1.25x faster than plain decoding on 2k-token prompts, 1.23x at 14k and 1.18x at 28k.
On 28k-token prompts it takes 68.4 % less time per token than DFlash in vLLM 0.30.0 ROCm.
One switch turns the recommended settings on: UNO_RECOMMENDED=1.
On NVIDIA, the same release decodes 1.69x faster than plain decoding one request at a time on one RTX 3090, and 1.38x with eight requests in flight.
There it serves 721 tokens per second with eight requests in flight against 679 for DFlash in vLLM 0.30.0.
github.com/brntech/vllm-uno/…
Built on IFM's Uno (@IFM_AI), running in vLLM (@vllm_project). Open source, free to run.
Release: github.com/brntech/vllm-uno/…
Image: ghcr.io/brntech/vllm-uno:0.4…
Adapter: huggingface.co/Broadnet/gemm…
The fastest Gemma 4 drafter now runs on AMD too.
On one Radeon AI PRO R9700 with Gemma 4 26B, Uno for vLLM v0.4.4 decodes 1.40x faster than plain decoding one request at a time, 1.44x with four requests in flight and 1.53x with eight.
Against DFlash in vLLM 0.30.0 ROCm, the fastest other drafter we tested, it takes 25.0 % less time per token one request at a time.
The output is the model's own.
Your system has the data. Your customer needs the update.
With BroadNet A2P Messaging, turn system activity into timely customer communication, from alerts and confirmations to reminders and important updates.
Close the gap. Keep customers informed.
📩 sales@broadnet.me
#BroadNet #A2PMessaging #CustomerCommunication #EnterpriseMessaging
Sent is a button. Delivered is a route. 🌍
Direct connections, premium routes, wholesale routes & SS7 connectivity—built for reliable global messaging.
Route smarter with BroadNet. broadnet.me
#SMSRouting #A2P #Telecom
Run the fastest Gemma 4 26B A4B on a single RTX 3090: ~215 tokens/sec on real production traffic.
New Uno adapter, trained on tiny 28.6k open prompts:
• 1.57x faster than plain decoding, now matching DFlash on short traffic
• Long prompts: 137 tok/s at 28k tokens, 2x DFlash (69) and well ahead of base (105)
• Lossless: same output distribution as the model on its own
Huge thanks to Google @googlegemma for the model and @IFM_AI for Uno.
Drop-in for Uno for vLLM (@vllm_project) v0.4.2, no new image.
huggingface.co/Broadnet/gemm…
#LLM #vLLM #Gemma
SMS. WhatsApp. Email. One connected conversation.
Keep every customer interaction together with Omnichannel.
sales@broadnet.me | broadnet.me
#Omnichannel #BusinessMessaging
Uno for vLLM v0.4.0 is out: the fastest Gemma 4 26B on long prompts we've measured.
@googlegemma Gemma 4 26B is the model BroadNet runs in production for our clients, in English and Arabic.
On one RTX 3090: 1.3x to 1.8x faster than DFlash, the fastest Gemma 4 drafter on this GPU, from 2k to 28k-token prompts.
1.52x faster than plain decoding on our real production traffic.
32k context. Lossless: the output is the model's own.
The drafter is a tiny adapter trained on 10k open prompts on a single GPU: 11x smaller than the MTP or DFlash drafters, trained on ~900x less data than the Uno paper's reference adapter.
Built with @GoogleAI model on @IFM_AI 's Uno, running in @vllm_project.
Here's what happens to speculative decoding when the prompt gets long.
@GoogleAI @googlegemma
Gemma 4 26B on one RTX 3090, 28k-token prompt:
DFlash: 0.66x. Slower than not speculating at all.
Uno: 1.30x. Still pulling ahead.
That's 2x DFlash. Lossless. Same card, same model, same AWQ 4-bit weights, faster than plain at every length from 2k to 28k.
The drafter behind it? A tiny 10k open prompts, trained on a single GPU. No cluster, no mountain of data.
Uno runs in @vllm_project today: github.com/brntech/vllm-uno
The long-context build and the new drafter land in the next release, with the @IFM_AI team.
Code and container: github.com/brntech/vllm-uno
Adapter: huggingface.co/Broadnet/gemm…
Release notes: github.com/brntech/vllm-uno/…
