#basemodel
🌌 And we done for the night! I thoroughly enjoyed that expedition, and I'm so excited for the future of No Man's Sky as well as whatever they have planned for Light No Fire.

Raided out to @struggleforwaffles.bsky.social tonight!

I'll be back on SUNDAY for more work on the Carnievore basemodel!
October 3, 2026 at 4:59 AM
i didnt play mhg (i did later get mhgu and not play it, but, such as it is) but its demo was free on 3ds and at the time my budget was limited to the cheapest basemodel version of a 2ds and whatever was available on the eshop for free. and i played the mhg demo a lot
September 29, 2026 at 5:56 AM
This weakly suggests that (for some classes of model) the labs do direct RL on exploit-finding capacity. It seems extremely likely that this was done for Mythos in specific

Other claim: yes, there's style transfer from the corpus to the basemodel. Basemodel training is like very lossy compression
September 19, 2026 at 4:42 PM
Good news, WSL is working now
Today I continued with FastAPI, using pydantic BaseModel to define a Student schema and set up routes to create and get student data. Also now running everything inside WSL Ubuntu.

What's one FastAPI feature you wish you knew earlier when you started?
September 18, 2026 at 1:52 PM
Anybody needs a #basemodel for #3Dmodeling? I don't know if it's any good cuz it's my 1st try at it, but here's a base silhouette I just created in #ibispaintx. Use it if you want. Customise it as you wish to fit.

Silhouette mostly so it can be used as ref or base for your own models

#blender #art
August 29, 2026 at 2:59 PM
Modeling three markets (BR / US / Crypto) with one rebalancing engine
> Part of a series on building **Balance** , a portfolio rebalancing app, as a solo developer. Balance started as a Brazilian-only app: B3 tickers, prices in reais, Brazilian income tax. Then came US stocks. Then crypto. The naïve path would have been three apps glued together — three price fetchers, three sets of forms, three rebalancing engines. Instead, the whole thing pivots on a single field: class Portfolio(BaseModel): market = models.CharField( max_length=6, choices=[('BR', 'Brasil'), ('US', 'Estados Unidos'), ('CRYPTO', 'Cripto')], default='BR', ) `market` is the one axis of variation. Everything market-specific branches on it — and crucially, the **rebalancing engine itself doesn't.** Let me show where the branches live and why the core stays untouched. ## Where the code actually differs Three markets differ in four concrete places: Concern | BR | US | Crypto ---|---|---|--- Price source | BRAPI → Yahoo (`.SA`) | Yahoo (direct) | Binance Currency symbol | R$ | $ | $ Tax rules | IR (20k exemption) | Capital Gains (FIFO) | IR cripto (35k) Quantity precision | whole shares | whole shares | 8 decimals Notice what's _not_ on that list: the rebalancing math, the category/target model, the deposit flow, the snapshots. That's ~90% of the domain, shared verbatim. ## Branch 1: price routing The price service picks a source from the portfolio's market. The trick is keeping the _interface_ identical — every path returns a `Decimal | None`: def _fetch_price(self, ticker: str, market: str) -> Decimal | None: if market == 'CRYPTO': return fetch_crypto_price(ticker) # Binance, no .SA if market == 'US': return self._fetch_yahoo(ticker) # AAPL, not AAPL.SA # BR: BRAPI first, Yahoo with .SA as fallback return self._fetch_brapi(ticker) or self._fetch_yahoo(f'{ticker}.SA') Callers never know which source answered. They get a price or a `None`, and the cache (`MarketPrice`) sits in front of all three. New market = new branch here, nothing else in the pipeline changes. ## Branch 2: market-aware forms A US ticker is `AAPL`. A Brazilian one is `PETR4`. A crypto one is `BTC`. The form adapts its validation, asset-class choices and currency labels from the portfolio's market: class AssetForm(forms.ModelForm): def __init__(self, *args, market='BR', **kwargs): super().__init__(*args, **kwargs) if market == 'US': self._ticker_re = r'^[A-Z]{1,6}$' self.fields['asset_class'].choices = STOCK_ETF_ONLY self.fields['current_price'].label = 'Price ($)' elif market == 'CRYPTO': self._ticker_re = r'^[A-Z]{2,10}$' self.fields['quantity'].widget.attrs['step'] = '0.00000001' else: # BR self._ticker_re = r'^[A-Z]{3,6}\d{0,2}$' Same `ModelForm`, three personalities. The view just passes `market=portfolio.market`. ## Branch 3: precision (the sharp edge) This is the one that bites. A stock quantity of `2` is fine in `decimal_places=2`. A crypto quantity of `0.00123456` silently rounds to `0.00`. The fix was to widen the precision everywhere money or quantity flows — `Asset.quantity`, `current_price`, `avg_cost`, `MarketPrice.price`, `Transaction` fields — all to `max_digits=20, decimal_places=8`. But that would over-format stocks ("2.00000000 shares"). So the _display_ branches, while the _storage_ doesn't: {% if is_crypto %} {{ asset.quantity|floatformat:"-8" }} {# 0.00123456 #} {% else %} {{ asset.quantity|floatformat:"0" }} {# 2 #} {% endif %} `floatformat:"-8"` strips trailing zeros, so `1.50000000` shows as `1.5`. Store wide, render narrow. ## Branch 4: the rebalancing engine _almost_ doesn't branch I said the engine is shared — that's 95% true. The one place it peeks at `market` is where whole-share assumptions break: if self.portfolio.market == 'CRYPTO': quantity = (budget / price).quantize(Decimal('0.00000001')) # fraction else: quantity = int(budget / price) # whole units That's it. Two or three of these guards across the whole service. Everything else — gap calculation, budget distribution, leftover spending — is market-agnostic because it works in _money_ , and money is the same shape in every market. ## The trade-off I'd flag The honest downside: `if market == 'CRYPTO'` sprinkled across the codebase is a smell that's tolerable at three markets and would rot at ten. If a fourth market with genuinely different mechanics showed up (say, options), I'd refactor those branches into a strategy object — `market.quantize(budget, price)` — rather than keep growing the conditionals. But premature abstraction has its own cost. At three markets, the branches are few, local, and obvious. Designing a pluggable "market backend" architecture on day one would have been more code to serve a flexibility I didn't yet need. The principle that held up: **model the axis of variation explicitly (`market`), branch where behavior genuinely differs, and keep the shared core working in the most general currency you have — money.** _Balance runs Brazilian, US and crypto portfolios through one rebalancing engine, with a consolidated dashboard that converts everything to a single currency. Link in profile. How do you handle multi-tenant-style variation in your apps — branching or strategy objects? Curious in the comments._
dev.to
August 25, 2026 at 11:07 PM
Modelando três mercados (BR / US / Cripto) com um só motor de rebalanceamento
> Parte de uma série sobre construir o **Balance** , um app de rebalanceamento de carteira, como dev solo. O Balance começou como um app só do Brasil: tickers da B3, preços em reais, IR brasileiro. Depois vieram ações dos EUA. Depois cripto. O caminho ingênuo teria sido três apps colados — três buscadores de preço, três conjuntos de formulários, três motores de rebalanceamento. Em vez disso, tudo gira em torno de um único campo: class Portfolio(BaseModel): market = models.CharField( max_length=6, choices=[('BR', 'Brasil'), ('US', 'Estados Unidos'), ('CRYPTO', 'Cripto')], default='BR', ) `market` é o único eixo de variação. Tudo que é específico de mercado ramifica nele — e, crucialmente, o **motor de rebalanceamento em si não ramifica.** Deixa eu mostrar onde os branches vivem e por que o núcleo fica intocado. ## Onde o código realmente difere Três mercados diferem em quatro pontos concretos: Aspecto | BR | US | Cripto ---|---|---|--- Fonte de preço | BRAPI → Yahoo (`.SA`) | Yahoo (direto) | Binance Símbolo de moeda | R$ | $ | $ Regras de IR | IR (isenção 20k) | Capital Gains (FIFO) | IR cripto (35k) Precisão de quantidade | ações inteiras | ações inteiras | 8 casas Repare no que _não_ está na lista: a matemática do rebalanceamento, o modelo de categoria/meta, o fluxo de aporte, os snapshots. Isso é ~90% do domínio, compartilhado idêntico. ## Branch 1: roteamento de preço O serviço de preço escolhe a fonte pelo mercado da carteira. O truque é manter a _interface_ idêntica — todo caminho retorna `Decimal | None`: def _fetch_price(self, ticker: str, market: str) -> Decimal | None: if market == 'CRYPTO': return fetch_crypto_price(ticker) # Binance, sem .SA if market == 'US': return self._fetch_yahoo(ticker) # AAPL, não AAPL.SA # BR: BRAPI primeiro, Yahoo com .SA como fallback return self._fetch_brapi(ticker) or self._fetch_yahoo(f'{ticker}.SA') Os callers nunca sabem qual fonte respondeu. Eles recebem um preço ou um `None`, e o cache (`MarketPrice`) fica na frente das três. Novo mercado = novo branch aqui, nada mais no pipeline muda. ## Branch 2: formulários conscientes do mercado Um ticker dos EUA é `AAPL`. Um brasileiro é `PETR4`. Um de cripto é `BTC`. O formulário adapta validação, choices de classe e labels de moeda a partir do mercado da carteira: class AssetForm(forms.ModelForm): def __init__(self, *args, market='BR', **kwargs): super().__init__(*args, **kwargs) if market == 'US': self._ticker_re = r'^[A-Z]{1,6}$' self.fields['asset_class'].choices = STOCK_ETF_ONLY self.fields['current_price'].label = 'Preço ($)' elif market == 'CRYPTO': self._ticker_re = r'^[A-Z]{2,10}$' self.fields['quantity'].widget.attrs['step'] = '0.00000001' else: # BR self._ticker_re = r'^[A-Z]{3,6}\d{0,2}$' O mesmo `ModelForm`, três personalidades. A view só passa `market=portfolio.market`. ## Branch 3: precisão (a aresta afiada) Esse é o que morde. Uma quantidade de ações `2` cabe em `decimal_places=2`. Uma quantidade de cripto `0.00123456` arredonda silenciosamente pra `0.00`. A correção foi alargar a precisão em todo lugar onde dinheiro ou quantidade passa — `Asset.quantity`, `current_price`, `avg_cost`, `MarketPrice.price`, campos de `Transaction` — tudo para `max_digits=20, decimal_places=8`. Mas isso formataria ações demais ("2.00000000 ações"). Então o _display_ ramifica, enquanto o _armazenamento_ não: {% if is_crypto %} {{ asset.quantity|floatformat:"-8" }} {# 0.00123456 #} {% else %} {{ asset.quantity|floatformat:"0" }} {# 2 #} {% endif %} `floatformat:"-8"` remove zeros à direita, então `1.50000000` vira `1.5`. Guarda largo, mostra estreito. ## Branch 4: o motor de rebalanceamento _quase_ não ramifica Eu disse que o motor é compartilhado — isso é 95% verdade. O único ponto em que ele olha o `market` é onde a suposição de "ações inteiras" quebra: if self.portfolio.market == 'CRYPTO': quantity = (budget / price).quantize(Decimal('0.00000001')) # fração else: quantity = int(budget / price) # unidades inteiras É isso. Dois ou três desses guardas no serviço inteiro. Todo o resto — cálculo de gap, distribuição de orçamento, gasto do troco — é agnóstico ao mercado porque trabalha em _dinheiro_ , e dinheiro tem a mesma forma em todo mercado. ## O trade-off que eu sinalizo A desvantagem honesta: `if market == 'CRYPTO'` espalhado pelo código é um cheiro tolerável com três mercados e que apodreceria com dez. Se um quarto mercado com mecânica genuinamente diferente aparecesse (digamos, opções), eu refatoraria esses branches num objeto de estratégia — `market.quantize(budget, price)` — em vez de seguir engordando os condicionais. Mas abstração prematura tem seu próprio custo. Com três mercados, os branches são poucos, locais e óbvios. Projetar uma arquitetura plugável de "backend de mercado" no dia 1 seria mais código pra servir uma flexibilidade que eu ainda não precisava. O princípio que se sustentou: **modele o eixo de variação explicitamente (`market`), ramifique onde o comportamento genuinamente difere, e mantenha o núcleo compartilhado trabalhando na moeda mais geral que você tem — dinheiro.** _O Balance roda carteiras brasileiras, americanas e cripto por um só motor de rebalanceamento, com dashboard consolidado que converte tudo numa moeda. Link no perfil. Como você lida com variação tipo multi-tenant nos seus apps — branch ou objeto de estratégia? Curioso nos comentários._
dev.to
August 25, 2026 at 11:07 PM
when tariffs were looming we got a basemodel whirlpool for like 500 dollars (Big White Fridge) and boy this thing is great. insulation has gotten so much thinner and more efficient that the inside of the fridge is a lot roomier than it would have been 20 years ago. and it gets cold as fuck
August 24, 2026 at 3:07 PM
LOL. Welches Basemodel in freier Wildbahn macht heute noch solche Fehler?
Ah, was für ein tolles Cover des neuen Vanilla Ice Albums ... besonders gut gefallen mir die Hauptfigur mit abgeschnittenem schwebenden Finger, der Breakdancer in Grün mit 3 Beinen und nem verschmolzenen Arm, aber auch der in der Mitte mit nur einem Fuß ist top!

(When you just don't care anymore)
August 13, 2026 at 1:05 PM
260 horsepower S60. Thats the basemodel.
February 26, 2025 at 2:02 AM
How to Build Human-in-the-Loop Plan-and-Execute AI Agents with Explicit User Approval Using LangGraph and Streamlit

app_code = r''' import os, json, uuid import streamlit as st from typing import TypedDict, List, Dict, Any, Optional from pydantic import BaseModel, Field from openai import OpenAI…
How to Build Human-in-the-Loop Plan-and-Execute AI Agents with Explicit User Approval Using LangGraph and Streamlit
app_code = r''' import os, json, uuid import streamlit as st from typing import TypedDict, List, Dict, Any, Optional from pydantic import BaseModel, Field from openai import OpenAI from langgraph.graph import StateGraph, START, END from langgraph.types import Command, interrupt from langgraph.checkpoint.memory import InMemorySaver def tool_search_flights(origin: str, destination: str, depart_date: str, return_date: str, budget_usd: int) -> Dict: options = [ {"airline": "SkyJet", "route": f"{origin}->{destination}", "depart": depart_date, "return": return_date, "price_usd": int(budget_usd*0.55)}, {"airline": "AeroBlue", "route": f"{origin}->{destination}", "depart": depart_date, "return": return_date, "price_usd": int(budget_usd*0.70)}, {"airline": "Nimbus Air", "route": f"{origin}->{destination}", "depart": depart_date, "return": return_date, "price_usd": int(budget_usd*0.62)}, ] options = sorted(options, key=lambda x: x["price_usd"]) return {"tool": "search_flights", "top_options": options[:2]} def tool_search_hotels(city: str, nights: int, budget_usd: int, preferences: List) -> Dict: base = max(60, int(budget_usd / max(nights, 1))) picks = [ {"name": "Central Boutique", "city": city, "nightly_usd": int(base*0.95), "notes": ["walkable", "great reviews"]}, {"name": "Riverside Stay", "city": city, "nightly_usd": int(base*0.80), "notes": ["quiet", "good value"]}, {"name": "Modern Loft Hotel", "city": city, "nightly_usd": int(base*1.10), "notes": ["new", "gym"]}, ] if "luxury" in : picks = sorted(picks, key=lambda x: -x["nightly_usd"]) else: picks = sorted(picks, key=lambda x: x["nightly_usd"]) return {"tool": "search_hotels", "top_options": picks[:2]} def tool_build_day_by_day(city: str, days: int, vibe: str) -> Dict: blocks = [] for d in range(1, days+1): blocks.append({ "day": d, "morning": f"{city}: coffee + a must-see landmark", "afternoon": f"{city}: {vibe} activity + local lunch", "evening": f"{city}: sunset spot + dinner + optional night walk" }) return {"tool": "draft_itinerary", "days": blocks} ''' …
nexttech-news.com
February 16, 2026 at 10:37 PM
Hope you all had a wonderful time during Furality this year! Also, if you were considering any of my base models, you have no need to stress, but as a reminder:

The sales is happening all the way until the 25th this month! #Furality #VRChat #VRChatAvatar #BaseModel
June 10, 2026 at 12:03 AM
You could still specify that with JSON. Of course not "manually" - you'd write the structure using, e.g., Pydantic BaseModel, and use Pydantic's schema-gen to produce the gnarly JSON that's horrible for us but just for LLMs
November 25, 2024 at 2:26 PM
Knight Basemodel... 👀

#art #gamedev #3dmodel #Commision
July 9, 2026 at 12:56 PM
You ship code your AI agent wrote. It reviewed itself. "Looks correct."

Three hours later: nothing works. Logs say 200 OK. But the API is dead.

Missing `from pydantic import BaseModel`. The /v1/health endpoint — deprecated, returning 200 with an empty body. Silent failure.

🧵 1/3
June 1, 2026 at 2:30 PM
hoje aprendi na prática a não misturar modelos pydantic com inputs dict, ou é tudo basemodel ou é tudo dict, senão vai dar problema 🥲
April 15, 2025 at 8:13 PM
Prompt of the Day:

Girl with red hair standing on a rock, overlooking the city of "Vienna". Oil Painting.

Steps: 25, CFG scale: 7, width: 768, height: 512, Seed: 55902588, Clip skip: 2, baseModel: SD1
January 10, 2024 at 7:44 PM
not saying that you _should_ do this

.... i am just saying that you _can_

....... in <40 lines that pass pylance strict

if I run it, it outputs:

{
"title": "The Matrix",
"release_year": 1999,
"actors": [
"Keanu Reeves",
"Laurence Fishburne",
"Carrie-Anne Moss"
]
}
January 20, 2025 at 1:53 AM
Yes, there are situations when it is useful. For instance, you can receive some enormous size JSON through the API and then instantiate multiple Pydantic typed objects of different classes passing _the same_ JSON to them (extra parameters are just ignored). Without inheritance of BaseModel you can't
December 3, 2024 at 12:54 AM
Reposting this but I just wanna say after I finish up a personal model im almost done with, imma continue with this VRC basemodel

I plan to make it natively compatible with the Killer Kobold vore system and try my hand at a detailed mouth sculpt

Lotsa chonk sliders too >:3
Lil playdoughy rn but HEHEHEHE
September 2, 2025 at 3:44 PM
trump got a gold phone coming out apparently frm the basemodel ova blackberry. 🤔

but first check the 'titan 2 elite'
May 26, 2026 at 5:38 PM
I’ve come to really appreciate the named tuple in python. Still not the most elegant implementation but it has just about all the behaviors I want without the baggage of dataclass, pydantic BaseModel, etc.
March 5, 2025 at 11:10 PM
ちょっとワカッタ!
「Select Models」を開いて、baseModel「Stable Diffusion v1.5」または「v2.1」を選ぶと馴染みあるポンコツな仕上がりになるみたいです。更にセッティング操作次第では画質もちょっとさがるかも……🤔
August 3, 2023 at 11:14 AM