هوش مصنوعی، یادگیری ماشین و یادگیری عمیق
موضوع اصلی کانال
این یک بلاگ شخصی با طرز تفکر شخصی هست.
Core Python : @PyHints
تلاشی هم در یادگیری Rust دارم که درحال داکیومنت شدن هم هست؛ اگر شماهم به این زبان علاقمند هستید join یادتون نره
Rust: @PyRust
Created
Between 1 March 2018 and 30 June 2021— estimated from Telegram’s id allocation, not measured. How this range is calculated.
5 measurements spanning 6 days, net +27. Dots are measurements; the straight line between them is drawn to join them, not to claim we know the path taken in between — snapshots are recorded only when a count changes, so gaps mean “no change observed”, never “interpolated”. The vertical axis spans 9,746–9,781 and does not start at zero.
Measurement log — every subscribers count we have recorded
Measured (UTC)
Subscribers
Change
12 Aug 2026, 21:58
9,777
+14
10 Aug 2026, 03:12
9,763
+8
7 Aug 2026, 07:55
9,755
+5
6 Aug 2026, 12:30
9,750
no change
6 Aug 2026, 12:25
9,750
first reading
Engagement
20 posts held, back to 30 May 2026 — the reader has not yet reached the start of this channel’s public history, so older posts may sit further back, unread. Read across 16 pagesof Telegram’s post history, 20 posts per page.
ERR · 30 days
19.2%
avg views ÷ 9,777 subscribers
Avg views / post
1,870
5 posts measured
Reaction rate
0.907%
reactions ÷ views · ER floor
Posts in window
5
of 20 held
ERR is average views per post over the last 30 days divided by subscribers, the definition TGStat uses, so this figure is comparable with the one you will see elsewhere. It falls structurally as a channel grows: a high ERR on a small channel and a low one on a large channel describe reach mathematics, not quality. We publish the figure and the sample it came from and pass no verdict on it.
ER is defined industry-wide as (forwards + reactions + comments) ÷ views— note the denominator is views, not subscribers. Telegram’s public web preview carries views and reactions but not forward or comment counts, so the reaction rate above is the reactions term only and is therefore a floor: the true ER for this channel is higher by an amount we have not measured and will not estimate.
What these figures were computed from
Window
Rolling 30 days · latest post in window 2 August 2026
Posts held
20 (30 May 2026 – 2 August 2026)
Views total
9,370
Reactions total
85
Forwards / comments
not exposed by the public surface — not measured, not estimated
Readings taken
13 Aug 2026, 05:52 UTC
Views are a single reading per post, taken at the time above. A post published in the last day or two is still accumulating views, which pulls the 30-day average down slightly. That is a property of the standard definition rather than a fault in it, so we keep the definition rather than “correcting” the number into something nobody can reproduce.
Precision. Telegram publishes view counts on its public widget in short form — 8.12K, 3.7M — so any reading at or above 1,000 reaches us rounded to three significant figures, and only counts below 1,000 are exact. Averages and rates derived from them are shown to the same precision rather than to the unit: a figure like 3,701,250 would assert digits nobody measured.
Reaction counts are published per emoji and rounded the same way, so a total below 1,000 is exact and a larger one is a sum that may carry a rounded component from each emoji above 1,000. Because it is a sum, it does not look rounded — read a large reaction total as three significant figures per contributing emoji rather than as the figure it prints.
What this channel posts
Photos
370
Videos
41
Links
692
Lifetime counters from Telegram’s own channel header, read 13 August 2026 — not the date at the top of this page, which is when the subscriber count was last read. Below Telegram’s rounding threshold, so these counts are exact.
Video runtime
6s
Average length
6s
Measured directly from 1 video with a duration reading, out of the posts we hold for this channel — not this channel’s whole posting history, only the sample this register has actually read. An exact reading to the second, taken from the post itself rather than from Telegram’s own rounded chrome, so it carries no ≈ mark.
Reaction mix
356 reactions across 20 posts, in 2 distinct kinds. The most used accounts for 56.2% of them.
Every reaction kind recorded on the sample, most used first
Reaction
Count
Share
Share, drawn
❤
200
56.2%
👍
156
43.8%
No sentiment is inferred, and none should be read in. This table is ordered by count and by nothing else. Emoji do not carry stable meaning across languages or communities — 🙏 is thanks in one channel and mourning in another — so we publish which ones were pressed and how often, and pass no judgement on what an audience meant by them.
Precision. Telegram publishes reaction counts per emoji and short-forms each one — 4.34K, 1.2M — so any single kind at or above 1,000 reaches us at three significant figures, and only counts below 1,000 are exact. The shares above are ratios of those figures and carry the same error. This is also why the total here can differ slightly from a reaction total printed elsewhere on the page: both are sums of the same rounded parts, taken over samples with different edges.
Coverage. Reactions were read on 20 of the 20 sampled posts in this sample. Summed by Telegram’s own count on each post — not by adding up the per-emoji breakdown above — those same posts carry 356reactions in total: the kind of figure the paragraph above means by “a reaction total printed elsewhere on the page”.
Measured over the 20 most recent posts we hold, published 30 May 2026 to 2 August 2026, using the newest reading held for each. Telegram Stars are excluded: they are a payment, not a reaction, and they have their own section.
در ادامه پروسههای باگ یابی من از شرکتهای هوش مصنوعی.
الان متوجه شدم چرا خیلی از بچهها میگن که فقط با Claude میتونند کد بزنند؛ من یک باگ جدید پیدا کردم.
Qwen, Grok, ChatGpt, GLM5.2
رو تست کردم؛ این مدلهای پارسر ضعیفتری دارند.
اول اینکه بنظر میاد همشون کد رو هم مثل markdown باهاش برخورد میکنند و پارس میکنند چرا ؟
حجم زیادی کد بهشون دادم با پرامپت سختگیرانه (خیلی بیادبی هست وگرنه میذاشتم) و همشون چون رو حالت سخ…
زده من اینکار رو کردم بعدش دیگه چه کارهایی میشه کرد باهاش ؟
هیچی تلاش کردی چندتاش رو بهم وصل کنی ببینی چی میده یا اصلا چطوری باید کد بزنی برای اتصال چندتاش به هم دیگه ؟
سریعتر میشه؛ کندتر میشه ؟
حالا bottleneck کجاس ؟
اگر مدل بزرگتر باشه چی میشه ؟
اصلا چطوری میشه چندتاش رو بهم وصل کرد (کارت شبکه که نداره)
مدل P4 که روی RiscV هست چطوری کار میکنه ؟
اگر همه این کارها رو کردی و جواب همش رو درست متوجه شدی یک سری به این…
لطفا ایشون رو برای همکاری استخدام کنید
AI developer runs 28.9-million-parameter model on $10 ESP32-S3 microcontroller — uses Google's Per-Layer Embeddings technique, stores table on 16MB Flash memory
اگر انقدر دیوونه هست که جنین کدی رو توی وقتهای خالی خودش بزنه؛ ببین توی کار چه کدهایی میزنه.
پینوشت:
خروجیش ۱ توکن در ثانیه هست مهم کاری هست که کرده و کدی که زده
Claude Leak
داستان چیه ؟ یک اتفاق تکراری دیگه.
هرکسی روی Calude چت خودش رو لینک Share براش گرفته چت بصورت کامل عمومی شده.
یک نفر هم اومده توی گوگل زده
site:claude.ai/share
و دیده بله گوگل تمام و کمال همرو ایندکس کرده.
که خب یعنی خیلی شرکتهای دیگه هم اینکار رو کردند و البته خیلی جاهای دیگه هم دیتاهای مهم رو خوندند.
خیلی از بیزینسها ابراز نگرانی کردند و گوگل نتایج رو حذف کرد و گفته میشه که کلاد هم باگ رو برطرف کرد…
این رو معرفی کنم
OKF (by google)
وقتی مدلها با کانتکستهای بالای 128K و حالا بیش از 1M استاندارد شدند واقعا هزینه توسعه و نگهداری سیستمهای RAG برام قابل درک نبود.
توی یک سری از ابزارهای چت شما میتونید مشخص کنید چت با چه پیامهایی از قبل مشخص شدهای شروع بشه و ماهم از همین تکنیک استفاده میکردیم.
مثلا برای نیروهای تازه وارد به شرکت؛ شماتیک دیتابیس رو توی چت اول میذاشتیم و نیرو همزمان با خوندن کدها اگر سوالی روی دیت…
مدلهای MoE خیلی جذابتر از مدلهای Dense هستند بنظرم چون میتونی یک مدل خیلی بزرگتر رو روی یک GPU خیلی کوچکتر با سرعت بالا اجرا کنی مثل :
https://telegram.me/per3onnel/263
یک سری افراد نشستند و همین کار رو برای GLM.5-2 با فشرده سازی و اپیتیمایز کردن بسیار انجام دادند بطوری که طبق ادعا خودشون مدل 774 میلیارد پارامتری رو روی سیستم با 25GB رم و بدون نیاز به GPU اجرا کردند (همه چیز روی C نوشته شده)
بطور خیلی ساده کاری…
یک پستی از دوست و همکار قدیمی توی لینکدین دیدم راجب یک کتابی که درحال نوشته شدن و تکمیل هست مخصوص کسانی که به
Deep Learning + Rust
علاقمند هستند گفتم اینجا هم به اشتراک بذارم
Linked In
شخصا درحال خالی کردن وقت برای خوندنش هستم چون موارد جالبی پوشش داده شده
مثلا اینکه بدون نیاز به سیستمعامل بتونید کار دیپلرنینگ انجام بدید چیزی نیست که روی فریمورکها و پایتون و ... آموزش داده بشه
LLM
قراره جای برنامه نویسهارو بگیره ؟
من که کدم رو + دستورالعمل ریفکتور دادم تا برام تمیز کنه و خروجی که توی تصویر هست
نصف باگهای مدلهای LLM رو من درآوردم باید بهم هزینه پرداخت کنند واقعا
پینوشت:
تازه ۳۵ دقیقه هم طول کشید. خودم میزدم ۲۰ دقیقهای تموم میشد.
Needle: We Distilled Gemini Tool Calling into a 26M Model
این مدل برای تخصصی برای tool calling ساخته شده؛ من جایگزین مدل ۴ میلیارد پارامتری کردم و همچنان به درستی از ابزارهایی که داشتم استفاده میکنه.
البته ابزارهای من ساده و قدیمی هست (خیلی وقت هست از این پروژه استفاده نکردم) اما بنظرم مدلی هست که ارزش تست کردن داشته باشه واقعا؛ اگر برای tool calling نیاز به مدل دارید یک تستی هم روی این انجام بدید اما بیشتر از too…
95% of enterprise AI deployments fail to deliver value
در یک حرکتی چندتا از این شرکتهای AI فروش داخلی (کاسبان تحریم) روی موجهای اخیر درحال تلاش برای خوروندن LLM به شرکتهای دیگر هستند. شرکتهایی که همینطوری بخاطر تحریم؛ جنگ؛ قطعی اینترنت و تعطیلی به زور سرپا موندند.
برای همین خواستم دوتا مورد رو یادآوری کنم:
اولین مورد گزارش MIT روی میزان سودآوری AI (که ۹۵٪ میزان بازگشت 0 داشتند)
legal.io
مورد دوم؛
تجربه شخصی …
نسل بعدی مدلها بنظرم خیلی بهتر خواهند شد :
Qwen AgentWorld
بنظرم این حرکت در راستای JEPA خواهد بود و این Gap بین LLM و JEPA رو برای مدتی میپوشونه تا نتایج مدلهای بر پایه JEPA خیلی بهتر بشه.
این ادعای بنچمارک یک مدل ۳ میلیارد پارامتری هست؛ تخصصی برای تسکهایی که نیاز به Thinking دارند.
و نتایجی نزدیک به مدلهای بیش از 300x بزرگتر
Hugging face
Paper
❤17👍3
Showing the 12 most recent of 20 posts we hold for @pytens. View and reaction counts are the latest single reading for each post, not a live figure, and a recent post is still accumulating both. A view count marked ≈ was rounded by Telegram before we ever saw it — t.me prints views in full below 1,000 and to three significant figures above, so ≈1,200,000 means somewhere between 1,150,000 and 1,249,999. Unmarked counts are exact. Text is reproduced from the public post preview and truncated for length.
Citation-graph rank
Citation-graph rank — 831,839 of 1,183,361entries in the measured graph. A weighted position computed from the forward and mention edges below — republished posts weigh more than named mentions — and recomputed periodically, over the whole graph. Published only as this ordinal position, never as a score: a position is a fact, and a score printed beside one channel’s name would read as a verdict this register does not make. The two counts beneath stay separate for the same reason mentions are never summed with forwards anywhere else on this page — a named-by count costs nothing to manufacture. The top 100 by this measure, or how it is computed.
Forward network
Republished by
Channels on the register that have forwarded this channel's posts into their own feed.
Built only from forwarded posts we have actually read, on both sides. Coverage is early and deliberately incomplete: a missing link means we have not read the post that would prove it, never that the relationship does not exist. Counts are distinct forwarded posts observed, so they only ever go up as we read more.
Mentions
Named by 1 registered channel — every channel on the register whose own posts have named this one, by its current username or any other username it currently holds, merged from two separately captured readings of the same fact so a namer caught by only one of them is not missed and a namer both caught is not counted twice. A username this channel has since dropped is not matched — that handle may belong to someone else now, and crediting today’s namer to yesterday’s owner would misattribute it.
Named by
Channels on the register whose posts name this channel's handle.
A mention is a weaker signal than a forward and is counted separately for that reason — naming a channel is not republishing it, and a handle in a post body is easy to place deliberately. The post counts beside each row below are distinct posts in which the handle appeared, from posts we have read on both sides — the “Named by N registered channels” figure above is a different count, of distinct NAMING CHANNELS rather than posts, and is not the sum of the rows under it.
Cite this entry
A live page changes as we take new readings, so a citation should name the measurement it is based on, not just the URL. The line below cites the subscriber count as measured 12 August 2026 — this
entry's latest reading, not the date you are reading this.
“دستاوردهای یادگیری عمیق(InTec)” (@pytens), 9,777 subscribers as measured 12 August 2026. Telegram Register, tgregister.com/channel/pytens.
Full measurement history, CC BY 4.0. Every reading this register holds for this entry, not just the latest one, as a dated, downloadable record: CSV · JSON. Free to use with attribution to tgregister.com. Each file carries its own generation timestamp, which is the figure to cite for exactly when the data was retrieved.
