ุงูุชุจู ูุฅูุณุงู โ Write Like a Human
๐ชถ
Cut stock phrases locally or rewrite text with an open LLM
Murtadha hamid PRO
murtadha9
AI & ML interests
Building applied AI systems: models, data pipelines, and monitoring & analysis tooling across text, speech, and vision โ with a focus on Arabic and the Iraqi dialect
Recent Activity
updated a Space about 9 hours ago
murtadha9/write-like-a-human updated a Space 2 days ago
murtadha9/match-audio-sound updated a Space 2 days ago
murtadha9/edit-video-by-typingOrganizations
None yet
updated a
Space about 9 hours ago
updated 7
Spaces 2 days ago
ุงุถุจุท ุงูุตูุช ุญุณุจ ู ุซุงู โ Match Audio Sound
๐
Match your track's sound to a reference example
ุนุฏูู ุงูููุฏูู ุจุงููุชุงุจุฉ โ Edit Video by Typing
๐
Edit a video by typing a request in Arabic or English
ุญููู ุตูุบุฉ ุงูุตูุฑุฉ โ Change Image Format
๐ช
Convert images between 7 formats, singly or in ZIP batches
ุตุบูุฑ ุญุฌู ู ูู PDF โ Shrink PDF Size
๐ชถ
Shrink PDF files while keeping Arabic text searchable
ุฏุจูุฌ ุงูููุฏูู ุฅูู ุงูุนุฑุจูุฉ โ Dub Video to Arabic
๐
Dub English videos into Arabic voice and subtitles
ุญููู ููุฑุชู ุฅูู ููุฏูู โ Topic to Short Video
๐
Arabic topic to vertical short video with voice & captions
ุฒูุจุฏุฉ โ Crux
๐ซ
Arabic-first extractive text summarizer
published 2
Spaces 4 days ago
posted an update 5 days ago
Post
72
๐ฌ ุงุฑูุน ููุฏูู ุฅูุฌููุฒูุงู.. ูุงุฎุฐู ู
ุฏุจูุฌุงู ุจุงูุนุฑุจูุฉ ุงููุตุญู ุฌุงูุฒุงู
ุงูุฃุฏุงุฉ ุชุชููู ุนูู:
๐ ุชูุฑูุบ ุงูููุงู ุจุชูููุชู
๐ ุชุฑุฌู ุฉ ุฅูู ุงููุตุญู
ูุดุงุดุฉ
ูููุงุชู ูุตู๐๏ธ ูุทู ุจุตูุช ุนุฑุจู
โฑ๏ธ ุถุจุท ุงูุตูุช ุนูู ุงูุชูููุช
๐ฌ ุชุฑุฌู ุฉ ุนุฑุจูุฉ ุนูู
ุงุญุชู ูุทูุงุจู.. ู ุญุชูู ุฅูุฌููุฒู ูุตู ุฅูู ุฌู ููุฑ ุนุฑุจู ุจุฏู ุฃู ุชูุฑูุบ ูุชุชุฑุฌู ูุชุณุฌูู ุตูุชู ูุชุถุจุท ุงูุชูููุช ุจูุฏู
ู ุฌุงูู ููุนู ู ู ู ุงูู ุชุตูุญ ููุง ูุญูุธ ู ููุงุชู.
๐ https://huggingface.co/spaces/murtadha9/dub-video-to
-arabic
ุงูุฃุฏุงุฉ ุชุชููู ุนูู:
๐ ุชูุฑูุบ ุงูููุงู ุจุชูููุชู
๐ ุชุฑุฌู ุฉ ุฅูู ุงููุตุญู
ูุดุงุดุฉ
ูููุงุชู ูุตู๐๏ธ ูุทู ุจุตูุช ุนุฑุจู
โฑ๏ธ ุถุจุท ุงูุตูุช ุนูู ุงูุชูููุช
๐ฌ ุชุฑุฌู ุฉ ุนุฑุจูุฉ ุนูู
ุงุญุชู ูุทูุงุจู.. ู ุญุชูู ุฅูุฌููุฒู ูุตู ุฅูู ุฌู ููุฑ ุนุฑุจู ุจุฏู ุฃู ุชูุฑูุบ ูุชุชุฑุฌู ูุชุณุฌูู ุตูุชู ูุชุถุจุท ุงูุชูููุช ุจูุฏู
ู ุฌุงูู ููุนู ู ู ู ุงูู ุชุตูุญ ููุง ูุญูุธ ู ููุงุชู.
๐ https://huggingface.co/spaces/murtadha9/dub-video-to
-arabic
published 5
Spaces 5 days ago
ุญููู ุตูุบุฉ ุงูุตูุฑุฉ โ Change Image Format
๐ช
Convert images between 7 formats, singly or in ZIP batches
ุตุบูุฑ ุญุฌู ู ูู PDF โ Shrink PDF Size
๐ชถ
Shrink PDF files while keeping Arabic text searchable
ุฏุจูุฌ ุงูููุฏูู ุฅูู ุงูุนุฑุจูุฉ โ Dub Video to Arabic
๐
Dub English videos into Arabic voice and subtitles
ุญููู ููุฑุชู ุฅูู ููุฏูู โ Topic to Short Video
๐
Arabic topic to vertical short video with voice & captions
ุงูุชุจู ูุฅูุณุงู โ Write Like a Human
๐ชถ
Cut stock phrases locally or rewrite text with an open LLM
reacted to sergiopaniego's post with ๐ฅ 7 days ago
Post
1,152
if you want to deep dive into how the new @MistralAI Large 4 was built, we sat down with the team to learn straight from them!
scaling, specialization, evals, cyber, sandboxing and why open weights matter
a really detailed look at what it takes to build and release an open frontier model, super recommended
https://www.youtube.com/watch?v=FczbSauVvjw
scaling, specialization, evals, cyber, sandboxing and why open weights matter
a really detailed look at what it takes to build and release an open frontier model, super recommended
https://www.youtube.com/watch?v=FczbSauVvjw
reacted to their post with ๐ฅ 8 days ago
Post
82
Glyph Lens
Instead of retyping printed text from a photo or scanned page word by word, you get it back in Arabic or English, ready to copy or download. It also highlights the words it may have misread right on the image, so you only need to check those. Free, runs in the browser, stores nothing.
๐ murtadha9/glyph-lens
Instead of retyping printed text from a photo or scanned page word by word, you get it back in Arabic or English, ready to copy or download. It also highlights the words it may have misread right on the image, so you only need to check those. Free, runs in the browser, stores nothing.
๐ murtadha9/glyph-lens
reacted to savoji's post with ๐ฅ 8 days ago
Post
1,819
One Qwen3-VL-8B checkpoint improves all three streaming spatial benchmarks at once: OST-Bench 50.52 to 63.93 (+13.41), UCS-Bench +7.44, OVO-S-Bench +4.91. The change is in the supervision, with no new memory module or annotation.
The ChronoReplay paper calls the failure mode interface lock-in: post-train on one answer format and that readout improves while others stall. A categorical-only control hits 58.03 on OVO-S but drops OST from 50.52 to 47.48.
๐ The recipe
โข Keep targets in native formats (option, metric value with units, timestamp)
โข Replay episodes in source order, each target attached to its causal prefix
โข Loss mass: 2/3 on 33,180 single-turn examples, 1/3 on 6,381 episodes (alpha = 0.5)
๐ The numbers
โข Three-benchmark mean 56.86, against 50.74 (categorical-only) and 50.76 (trajectory-only)
โข Chronological order beats a shifted-order control by about 2.41 OST points on two seeds; the 13.41 belongs to the whole recipe
โข Adaptive frame selection adds +4.09 on OVO-S (57.98) with 80.50 frames instead of 128
โ ๏ธ Caveats
โข One backbone, indoor scenes only; the single-family comparison is confounded by data coverage
โข The repo has a CPU smoke test and a --dry-run training command, but no license and no fine-tuned weights
ChronoReplay: One Answer Interface Is Not Enough for Streaming Spatial Intelligence (2610.34283)
https://arxiv.org/abs/2610.34283
https://github.com/yihaohu0118/ChronoReplay
If you fine-tune VLMs for embodied work, do you score categorical, metric and temporal readouts together?
๐ Read it elsewhere
โข The deep dive on Medium: https://medium.com/p/8d2fb1bf832d
โข The long read on Substack: https://sajjadpakdamansavoji.substack.com/p/chronoreplay-lifts-qwen3-vl-8b-by
โข The thread on X: https://x.com/saavoji/status/2116229682844877156
โข Every version on sajjadpsavoji.github.io: https://sajjadpsavoji.github.io/blog#chronoreplay
The ChronoReplay paper calls the failure mode interface lock-in: post-train on one answer format and that readout improves while others stall. A categorical-only control hits 58.03 on OVO-S but drops OST from 50.52 to 47.48.
๐ The recipe
โข Keep targets in native formats (option, metric value with units, timestamp)
โข Replay episodes in source order, each target attached to its causal prefix
โข Loss mass: 2/3 on 33,180 single-turn examples, 1/3 on 6,381 episodes (alpha = 0.5)
๐ The numbers
โข Three-benchmark mean 56.86, against 50.74 (categorical-only) and 50.76 (trajectory-only)
โข Chronological order beats a shifted-order control by about 2.41 OST points on two seeds; the 13.41 belongs to the whole recipe
โข Adaptive frame selection adds +4.09 on OVO-S (57.98) with 80.50 frames instead of 128
โ ๏ธ Caveats
โข One backbone, indoor scenes only; the single-family comparison is confounded by data coverage
โข The repo has a CPU smoke test and a --dry-run training command, but no license and no fine-tuned weights
ChronoReplay: One Answer Interface Is Not Enough for Streaming Spatial Intelligence (2610.34283)
https://arxiv.org/abs/2610.34283
https://github.com/yihaohu0118/ChronoReplay
If you fine-tune VLMs for embodied work, do you score categorical, metric and temporal readouts together?
๐ Read it elsewhere
โข The deep dive on Medium: https://medium.com/p/8d2fb1bf832d
โข The long read on Substack: https://sajjadpakdamansavoji.substack.com/p/chronoreplay-lifts-qwen3-vl-8b-by
โข The thread on X: https://x.com/saavoji/status/2116229682844877156
โข Every version on sajjadpsavoji.github.io: https://sajjadpsavoji.github.io/blog#chronoreplay
- 2 replies
reacted to crp4222's post with ๐ฅ 8 days ago
Post
2,740
I trained a small PII detector for PrivAiTe, my local proxy that masks personal data before a prompt reaches an LLM: crp4222/privaite-pii-minilm.
It is a multilingual MiniLM (117M parameters, 15 types from names and contacts to IDs, bank details and secrets), trained only on datasets whose licenses allow commercial use, released under BSD 3-Clause, with PyTorch, ONNX fp32 and int8, and Core ML builds.
On 9 public test sets it never saw, it masks as many person names as the best of GLiNER PII, NVIDIA GLiNER-PII, GLiNER2, Piiranha and OpenAI privacy-filter, or more, on 6 of them, usually with higher precision. It trails GLiNER PII on short lowercase voice requests and NVIDIA on court rulings.
It is also 2.5 to 7 times smaller and much faster: 9 ms per 1,000 characters on an M1 Pro Neural Engine and 23 ms on CPU in int8, against 186 to 621 ms for the others on the same machine and documents. Numbers, test sets and limits are on the card: crp4222/privaite-pii-minilm
It is a multilingual MiniLM (117M parameters, 15 types from names and contacts to IDs, bank details and secrets), trained only on datasets whose licenses allow commercial use, released under BSD 3-Clause, with PyTorch, ONNX fp32 and int8, and Core ML builds.
On 9 public test sets it never saw, it masks as many person names as the best of GLiNER PII, NVIDIA GLiNER-PII, GLiNER2, Piiranha and OpenAI privacy-filter, or more, on 6 of them, usually with higher precision. It trails GLiNER PII on short lowercase voice requests and NVIDIA on court rulings.
It is also 2.5 to 7 times smaller and much faster: 9 ms per 1,000 characters on an M1 Pro Neural Engine and 23 ms on CPU in int8, against 186 to 621 ms for the others on the same machine and documents. Numbers, test sets and limits are on the card: crp4222/privaite-pii-minilm
- 5 replies