put AI declaration, and got scammed 3 times by different websites, claiming they give free GPU hours but ended up giving nothing but takng my email. currently working on a way to host my model, working on the exact workflow that my next AI project under thalamus will use, it will be JARVIS so here is a spoiler. it will beat fable 5. and I have ran out of the weekly Claude limit on all my 5 email IDs.
the instaler is still ended up being corrupt or incorrect. neither me nor AI can understand the problem here. and today i drank 3 cups of coffee instead f my usual 2.

here is a screenshot of the error im recieving.
and now i will send my shit for recertification again, now I have everything in place, including the AI declarations. hopefully, i wont hve to do it for the fifth or sixth time.

i will be surprised if i Dont get a migrane tommorow morning cause ive been spamming the .exe to work, the build for the .exe installer and .msi installer is failing again and again.
to the shipwrights, the .exe is completely safe and virus free, there are chances that you will install a .msi instead of .exe
on running, there are chances that the installer will fail, if it does, I know that it is failing and I am trying to fix it, if it works, then it works.

heres a screenshot of the build failing.
all I hope is that my ship gets verified this tme, 4rth time sending this bad boy for re cert.

for the Ship reviewing team
sry for last re cert request, forgot that I have to reqrite m readme.md
did that, now im requesting re cert again. hopeefuylly it passes this time.
ps.my credits request is still pending, AMD relied that they will respond ASAP, it is under review(do companes do that) and then also sent an apology letter for the delay(why? why did they send it?).
and please remove the compulsion of having a photo/ss for software projects.

I will now request Recertification, I think that I have done enough.

To the reviewers, there are chances that you will get to use my self trained model if my AMD clloud credits request gets approved or my Google Cloud Instance quota gets increased, otherwise, the only change I have made is nothing. Im unable to put my AI online.
SS uploaded is my kaggle account o you can see proof of AI model.

devlog: the 100-hour milestone, a humaneval script blowout, and parsing raw logs

context recap (the architecture so far)

for anyone catching up on the architecture overhaul: my initial ship for thalamus got rejected because the UI looked like an AI wrapper template. instead of shifting pixels, i decided to completely redo the backend engine. i’m training a custom, specialized 32B model optimized explicitly for programming logic—named thalamus-thor-1. general fluff is scrubbed to save weight; a gemini researcher agent sub-team handles web scraping to cover info blind spots.

resource-wise, it’s a tight squeeze. puter.js failed on orchestration, so i’m entirely dependent on a drying runway of AWS credits while waiting on google cloud and cheap amd gpu server approvals. with school starting in two days, i’ve also been preparing a lightweight remote framework to monitor logs and trigger test suites from school lab computers during short periods. earlier tonight, i managed to bypass kaggle’s aggressive session killers, successfully compiled the FP8/4-bit compressed weights on their tpu clusters, and immediately threw the model straight into openai’s standard 164-problem HumanEval benchmark.

here is the live update from late tonight.

---

milestone: 100 hours in the dirt

first off, a quick personal timestamp: as of setting up these evaluation scripts, i have officially crossed the 100-hour mark of active development on this project. hitting triple digits while drowning in memory leaks, credit run-outs, and hardware timeouts is a brutal milestone, but the core engine is standing.

the humaneval script blowout

i watched the model successfully plow past problem 90, and it kept charging all the way through the end of the execution run. but right at the finish line, instead of outputting the final pass@1 score summary, the script threw a fatal calculation error and crashed out.

the evaluation script generated the completions, ran the test assertions, and executed the functional verification, but the section of the code responsible for processing the final combination formula or compiling the macro-averaged percentage broke completely. standard human-eval metric scripts use an unbiased estimator calculation to evaluate pass@k:

$$\text{pass@k} = 1 - \frac{\binom{n-c}{k}}{\binom{n}{k}}$$

if the array containing the total number of generated completions ($n$) or the correct passes ($c$) contains a single malformed row, a zero-division error, or an unhandled null exception from an execution timeout, the entire summary wrapper explodes. i’m staring at a broken metrics calculator, but the actual data isn’t lost.

shifting to manual validation

i’m not letting a broken parsing script mask the actual performance metrics of the weights after 100 hours of grind. the script managed to compile and save the fine-grained data out to the raw <input_path>_results.jsonl log file before the summary function choked. each row in that file contains the literal execution results marked cleanly as "passed", "failed", or "timed out".

instead of trying to debug the evaluator framework’s entry points tonight, i’m going to pull the raw JSON lines into a basic python script i’m writing right now. i’ll manually iterate through the rows, count the hard instances of "passed" across the tasks, and divide it by 164 to find the true, absolute pass@1 score myself.

the raw logs are sitting in the directory. writing the cleanup parser now to see what thalamus-thor-1 actually scored.

devlog: impatient engineering, skipping the wait, and running humaneval live

the project recap

quick context if you’re catching up: my initial submission for thalamus got rejected because the UI looked like a boilerplate wrapper. instead of a superficial skin-deep fix, i decided to overhaul the entire backend engine and train a custom, hyper-specialized 32B model optimized strictly for heavy coding logic—named thalamus-thor-1. conversational filler was stripped entirely from the datasets to keep the weights lean; my existing gemini agent team handles live web scraping to cover general knowledge blind spots.

the resource situation is a major scramble. puter.js was too limited for my 9-agent orchestration, so i abandoned it entirely. i’m burning through the absolute last of my AWS credits while waiting on approvals for google cloud and cheap amd gpu clusters. plus, school starts in a couple days, so i’m building a lightweight remote execution layer to trigger test suites and monitor terminal logs from the school computer lab during our free periods. earlier today, after a brutal 13-hour marathon fighting out-of-memory (OOM) errors and kaggle’s session killers on deepspeed ZeRO-3 initialization spikes, i finally pushed the parameters under the threshold and successfully compiled the weights on their tpu v5e-8 kernels.

here is the live update from 1.5 hours later.

---

the midnight deployment: saying screw it

the plan from my last post was to sleep on the weights and stand up the evaluation harness tomorrow morning. i lasted twenty minutes. when you spend 13 straight hours fighting initialization crashes just to get a model cooked, you don’t leave the weights sitting cold on a drive. the curiosity and momentum make it impossible to step away.

i went ahead, pulled down the evaluation scripts, configured the isolation wrappers, and slammed the model directly into the testing execution loop.

right now, thalamus-thor-1 is running a live trial by fire. it is chewing through openai’s standard HumanEval dataset in real-time. as of typing this, the model has already plowed past 90 out of the 164 total programming problems in the suite.

the screenshot attached is the raw stdout window of my evaluation terminal. you can see the script compiling generations, feeding function signatures, and executing validation unit tests back-to-back.

what the logs are proving

seeing the terminal scroll through 90+ consecutive problems without dropping a syntax exception or throwing infinite token loops is an immense relief. when you compress a 32B model down to tight FP8 and 4-bit configurations while scaling the context matrix toward an 800K target using RoPE and YaRN embeddings, the mathematical margin for error is razor-thin. one wrong scale factor and the model outputs broken brackets or repeats strings indefinitely.

instead, the model is executing functional correctness cleanly. it’s handling complex array manipulations, algorithmic logic, and nested structural filters completely raw. because the weights are completely unburdened by chat filler, it behaves less like a chatbot and more like a lightning-fast logic compiler—reading the docstring and immediately piping clean python code directly to the testing harness.

once it clears problem 164, the system will output the final pass@1 percentage. that single number will tell the definitive truth about the token economics i calculated earlier. if the pass rate holds up against these unit tests, we are looking at a model that can genuinely compete with sonnet 4.6 on pure engineering tasks, while operating at an estimated internal compute cost of just $2 per million input tokens and $6 per million output tokens—giving my agent pipeline high-tier logic execution at the price of haiku 4.5.

i’m staying up until the script hits 164. final data drops the second the testing loops finish executing.

devlog: thalamus-thor-1 is trained. recap, kaggle breakthrough, and humaneval

context recap

if you haven’t been tracking: my first ship attempt for thalamus got rejected because the UI looked like a generic boilerplate wrapper. instead of a superficial patch, i decided to completely overhaul the core engine and train a custom 32B model optimized strictly for raw code logic—named thalamus-thor-1. general knowledge was stripped out during data prep to keep weights lean; my existing gemini-led researcher team handles live web scraping instead.

infrastructure-wise, it’s a massive scramble. puter.js proved way too limited for my 9-agent pipeline orchestration, so i’m entirely stranded there, burning through the absolute last of my aws credits and waiting on gcp/amd approvals. also, school starts in two days, so i’m building a lightweight remote pipeline to trigger test suites and monitor logs straight from school lab machines during free periods.

---

the kaggle breakthrough

yesterday, i was stuck in deepspeed ZeRO-3 initialization hell. every time the notebook tried to load the base parameters on kaggle’s free tpu v5e-8 clusters, a massive compute spike would trigger their aggressive backend session killers, immediately dropping the kernel.

i spent hours adjusting chunk prefills, gradient accumulation steps, and stripping background overhead until the initialization sequence squeezed just under the termination threshold. the run finally executed, completed its QLoRA fine-tuning passes at FP8, scaled the context toward an 800K target using RoPE and YaRN with 4-bit compression, and successfully compiled the weights.

the screenshot attached to this update is the actual kaggle notebook page of my AI training notebook showing the setup.

initial sanity checks

compiling weights is only half the battle. when you aggressively compress a 32B model down to FP8/4-bit formats while applying long-context scaling embeddings, the positional math can warp. if it breaks, the model turns into a hallucination engine.

i spun up a local instance to run direct logic checks, and the structural syntax held together perfectly. it didn’t choke on deeply nested loops, accurately referenced parameters defined deep inside large codebase contexts, and skipped conversational fluff entirely to jump straight into code execution.

tomorrow’s trial: the humaneval protocol

vibe checks aren’t enough. tomorrow morning, i am setting up an automated evaluation harness to run thalamus-thor-1 through the full OpenAI HumanEval dataset (164 coding problems) to calculate the raw pass@1 rate.

this will prove the token economics math i did earlier: whether this custom setup can genuinely compete with sonnet 4.6 and opus 4.6 on pure engineering tasks while running at an estimated cost of $2/1M input and $6/1M output tokens (top-tier intelligence at the price of haiku 4.5).

hosting is still the looming issue. permanently serving this with an extended context window requires a dedicated 80gb VRAM prompt server configuration, which i’m trying to source while my teammate builds out the social media presence to push links.

on a side note, fighting package conflicts during this nightmare inspired a lightweight side-utility: a clean CLI tool for coders that natively resolves dependency hell. the core engine is working perfectly, and you can track it here: https://stardance.hackclub.com/projects/6323

evaluation metrics drop tomorrow morning. let’s see what the numbers show.

quick recap on why the scope completely escalated: my first ship attempt for thalamus got rejected because the UI looked like a generic template wrapper and the documentation felt AI-generated. instead of just patching the CSS, i decided to completely overhaul the core engine and train a custom 32B model optimized for raw code logic—named thalamus-thor-1.

here is the brutal reality of the last 24 hours:

1. the infrastructure and credit crunch
i previously planned to use puter.js as a temporary backend fallback router for the claude models while training. scratch that. puter.js is way too limited for the multi-agent architecture i’m running, so i’m entirely stranded there. right now, i’m burning through the absolute last of my AWS credits. i’ve applied for google cloud credits to spin up proper GPU instances, but everything hinges on getting that approval before the AWS runway hits zero. surviving on nothing but hope and coffee.

2. 13 hours battling kaggle
i spent 13 hours straight fighting out-of-memory (OOM) errors and kaggle’s infrastructure quirks. i successfully patched the local OOM issues, but now i’m hitting a wall with their session killers. the platform completely terminates the notebook due to compute spikes right when ZeRO-3 is about to initialize. the screenshot for this post is the raw kaggle output cell showing exactly where the infra drops the hammer.

because of this, i’ve applied for amd gpu credits. they’re vastly cheaper than nvidia options but offer massive amounts of vram, which is exactly what a high-context 32b setup needs to handle codebase ingestions via RoPE and YaRN.

3. Social Media
on the startup side, my teammate is taking over the social media presence and will start pushing updates the second the app is not hope it’s functional.

4. dependency utility side-project
ironically, fighting package conflicts during this 13-hour nightmare gave me an idea for a quality-of-life tool. i built a clean CLI tool for coders that natively resolves dependency hell. the core logic is already up and running perfectly.

you can track and follow the new CLI project here: https://stardance.hackclub.com/projects/6323

going back to tweaking the zero-3 configurations before the credits completely empty out.

Just saw that puter.js brand themselves as unlimited AI, they arrent actually unlimited.
Now on my hunt to find something to temporarily replace Puter.js for claude requests.

Gotta start training my AI fast.

finally back from my trip. i didn’t post the last update because the entire scope of what i’m doing just completely escalated.

to recap: my first ship attempt for thalamus got rejected. the feedback was that the UI looked too much like a generic AI wrapper boilerplate, the README felt AI-generated, and it needed unique features. instead of just doing a superficial skin change or tweaking the CSS to look pretty, i am using this as a reason to completely redo the underlying engineering and turn this into a true L4.5 or L5 agent that destroys things like claude code.

i attached a photo of my handwritten notes to this post. it maps out the exact problem statements i’m tackling and the major flaws of currently existing solutions out there. my handwriting is messy, but the technical breakdown is all there.

here is the actual game plan for the AI routing and training infrastructure:

right now, i am updating the app to route claude models via puter.js. to be clear, this isn’t for the user interface—puter.js is strictly my temporary backend AI router. it acts as a fallback bridge while i build out my own custom models. i’m going to benchmark my self-trained models against every single task in my 9-agent pipeline. until my model is 100% foolproof on a specific task, puter.js will keep routing that specific agent’s requests to claude. but the second my custom model qualifies and beats the benchmarks for code mode, puter.js stops getting requests for it and the pipeline switches over entirely to my model.

the actual training project is running parallel to this:

• i’m fine-tuning a 32B model optimized explicitly for pure coding and logical tasks.
• training is happening via QLoRA at FP8 with a 4k/8k base context, but i’m extending the context window all the way up to 800K using RoPE and YaRN with 4-bit compression, flashattention-3, and chunked prefills so it can digest entire codebases at once.
• i’m doing the actual training on kaggle kernels using the 30 free weekly hours on their TPU v5e-8 clusters.

based on my calculations, this model will completely fail at things like history or literature, but it doesn’t need to be good at them. my gemini researcher agent team already handles live web search and pulls fresh information for those blind spots anyway. on pure code, this thing should easily beat sonnet 4.6 and compete with opus 4.6.

the token economics are insane if this works: about $2 per million input tokens and $6 per million output tokens based on max output possible per hour versus GPU compute costs. that gives me sonnet/opus level intelligence at the price of haiku 4.5.

the real bottleneck right now isn’t the data or the training—it’s hosting. deploying a 32B model with an extended context window means i need to secure an 80gb VRAM per parellel prompt configuration. figuring out how to source or fund that infrastructure without incinerating cash is the main issue i’m working through while compiling the training scripts.
Today i will just do the rerouting to puter.js, then start working on fine tuning the 32b model. Wish me luck.

Following the rejection of my initial shipping submission, I am pivoting to address the required refinements. I am viewing this as a system audit rather than a failure—the core engineering is stable, but the interface and documentation need to move closer to a production-grade tool.

---

Engineering Status

Module Current Status Notes 9-Agent Pipeline Stable Successfully debugged AgentRouter integration. VM Bridge Alpha QEMU bridge is operational; optimizing latency. Claude Code Workflow Ready Pipeline errors resolved; full parsing functional.

---

Feedback Audit

The review team highlighted specific areas for improvement:

• Interface: The current UI requires a shift from “AI-template” style to a proprietary “Control Center” aesthetic.
• Documentation: The README is currently AI-generated; I will rewrite it to reflect the technical architecture of the 9-agent pipeline and the VM bridge implementation.
• Utility: Implementing “startup-specific” features that move beyond a standard chat wrapper.

---

Schedule & Roadmap

I am entering a maintenance window due to travel. No commits will be pushed during this period.

• June 20: Finalizing technical walkthrough assets.
• June 21 - 25: OFFLINE (Maintenance Mode).
• June 26: Resuming development. Priority: UI Refactor & Documentation Audit.

Project Repository: https://github.com/hardcoregamingsyle/thalamus

ps. for anyone wondering, use markdown to format your devposts

today has to be my most unproductive day yet. I am vey frustrated because of some error, and when I give it to claude code, guess what, Claude code is getting errors, Like it is topping in between, or giving me error 400, 401 or 403, That is what Im doing right now. it is just not working. Like I was fixing the Agent Output parasing so the Commands are Read properly. Thought to refractor all big files, Now refractoring myself would be time consuming, so I went to claude code, and giess what, Claude code is giving me errors, And i ended up spending the same number of time working as if I wouldve done manually. but atleast my code is fine and does not have any bugs.
Once the WebUI is fixed PC, I will work on WebUI for other Devices, then work on the .exe file for the VM for any OS, I havegathered the ISOs for Windows 10, Widnows 11(Both are Pro Vesions and Preactivated), MacOS 11to MacOS 16, Android 14 and Android 13, Linux, Ubuntu and other Open Source OS systems, aftre that I will work on iOS. which is byfar the most difficult one. I also have to make some changes to the Referal System and Token Deduction system, It is currently Highly overpriced.

Will update yall once I make some more progress

This is devlog so I dont think we should only send our app progressions but treat it like a communuty, so to start with it. I finally got rid of a 400 error, found out while getting rid of it, my codebase got infected with corrupted files, I am now using claude code to clean uo my codebase. also sadly, I had to delete my entir git history, which is very sad. I am now from 68 coomits to just 4(the first commit has +~70k lines as proof that i had real work). Anyways, heres a photo of 2 corrupted files which I manually found. I better get some sleep before I get a headache and weaken my eyes.

Today, I finally got the code mode and study mode to work, They are a bit slow, Im working on it, also. right now, Im busy debugging this error that Im getting by claude code, all help will be highly appriciated. My Project is ready to be shipped, Will ship it today, all that I need to do is improve the already good documentation.

REQUEST: Please help me solve this error that I am again and again getting while using claude code, I changed the settings.json of the .claude folder so I can use Claude code but the API will be of Agentrouter.org so I dont hve to buy the claude Plans. It works usually but sometimes I get this error, and when I di get it, it is just annoying.

Demo: https://thalamus.aphantic.skinticals.com
You will have to log in for More usage and access to Code mode and Research mode, had to restrict because it costs me money whenever someone uses it.

Well because my origional vibe coder got aquired by freebuff, I spent the rest of this time migrating from vly to Claude Code, made some progress, In initial testing, I found out that code mode and study mode were broken, Working on that, also stepping up my .exe from a Electron wrapper to a natice C++, C# Desktop app, Like of Minecraft/Roblox. This is taking me some time as claude code is kinda dumb for me because im like using it as my main Agent for the first time so it will take me some time getting the hang of it, Also I have cnnected Claude code to agentrouter, which is a third party that gives $100 initial credits and then $25 a day for signing in. Using my fav multi account system to get the most of this also. Will post another devlog as soon as I make some more progress. The image uploaded is just me working on code mode.

fixing the sandbox to actually work