Success story with MiMo-V2.5-GGUF:UD-Q5_K_XL
Mirrored from r/LocalLLaMA for archival readability. Support the source by reading on the original site.
I don't see many stories about this model, but after several attempts (after I finished finally reconfiguring my cluster) I did something useful with it: it wrote a built-in llama.cpp tool for executing C++ code and using the results.
Here's an exercise that MiMo V2.5 gave me to practice my skills with an RPN calculator (I didn't have one of those as a kid) and for which I asked MiMo again to write a program:
write a c++ program to calculate the compound interest for this problem: Problem 1: Calculating Compound Interest Scenario: You're writing a financial application and need to verify compound interest calculations. Problem: Calculate the future value of $10,000 invested at 5% annual interest compounded quarterly for 8 years using the formula: A=P(1+rn)^nt Where: P=10000 (principal) r=0.05 (annual interest rate) n=4 (compounding periods per year) t=8 (years) This is what the arguments for the tool call look like:
{ "code": "#include <iostream>\n#include <cmath>\n#include <iomanip>\n\nint main() {\n // Given variables\n double P = 10000.0; // Principal amount\n double r = 0.05; // Annual interest rate (5%)\n int n = 4; // Compounding periods per year (quarterly)\n int t = 8; // Time in years\n\n // Compound Interest Formula: A = P(1 + )^(nt)\n double amount = P * std::pow((1 + (r / n)), (n * t));\n double interest = amount - P;\n\n // Output results\n std::cout << std::fixed << std::setprecision(2);\n std::cout << \"Principal Amount: $\" << P << std::endl;\n std::cout << \"Annual Interest Rate: \" << (r * 100) << \"%\" << std::endl;\n std::cout << \"Compounding Periods per Year: \" << n << std::endl;\n std::cout << \"Time Period: \" << t << \" years\" << std::endl;\n std::cout << \"-----------------------------------\" << std::endl;\n std::cout << \"Future Value (A): $\" << amount << std::endl;\n std::cout << \"Total Interest Earned: $\" << interest << std::endl;\n\n return 0;\n}\n" } A bit better formatted:
#include <iostream> #include <cmath> // Required for std::pow #include <iomanip> // Required for std::setprecision int main() { // Variable definitions based on the problem statement double P = 10000.0; // Principal amount double r = 0.05; // Annual interest rate (5%) int n = 4; // Compounding periods per year (quarterly) int t = 8; // Time in years // Compound Interest Formula: A = P(1 + )^(nt) // std::pow(base, exponent) is used to handle the power calculation double amount = P * std::pow((1 + (r / n)), (n * t)); double interest = amount - P; // Formatting output to show 2 decimal places for currency std::cout << std::fixed << std::setprecision(2); std::cout << "Principal Amount: $" << P << std::endl; std::cout << "Annual Interest Rate: " << (r * 100) << "%" << std::endl; std::cout << "Compounding Periods per Year: " << n << std::endl; std::cout << "Time Period: " << t << " years" << std::endl; std::cout << "-----------------------------------" << std::endl; std::cout << "Future Value (A): $" << amount << std::endl; std::cout << "Total Interest Earned: $" << interest << std::endl; return 0; } I'll skip the code explanations the LLM wrote about the code (it's a bit hard to get them formatted on reddit) and show the output of the program:
Principal Amount: $10000.00 Annual Interest Rate: 5.00% Compounding Periods per Year: 4 Time Period: 8 years ----------------------------------- Future Value (A): $14881.31 Total Interest Earned: $4881.31 Which was almost the exact value I got on my calculator. ^^
How I did it: I used opencode and instructed the model to read tools/server/server-tools.cpp and then to implement a tool for compiling a program and getting the results. I did a few spelling mistakes in my prompts and there was a compiling mistake from a previous experiment, but then everything worked.
I'm a vibe coder now. Make your requests and I can try to merge them upstream 🦌
Edit1: fixed the expression to for the compound interest to include ^
[link] [comments]
More from r/LocalLLaMA
-
Been running Qwen3.6-27B through a 3-critic harness. The harness matters more than I thought
Jun 30
-
I Hate Dario Amodei, and everything he stands for.
Jun 29
-
Introducing LongCat-2.0 - , a large-scale MoE language model with 1.6 trillion total parameters and ~48 billion activated per token. This was the stealth model that was on Openrouter under the name 'owl-alpha'.
Jun 29
-
Krea-2-Turbo Image Model - Easy to be fully uncensored, but it can also EDIT Images!
Jun 29
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.