Icodetools commited on
Commit
794daf2
·
verified ·
1 Parent(s): ab222ac

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +19 -1
README.md CHANGED
@@ -32,6 +32,15 @@ tags:
32
  - vulcan
33
  ---
34
 
 
 
 
 
 
 
 
 
 
35
  # Vulcan-1-Coder (GGUF)
36
 
37
  A 2.5B coding and reasoning model built on **[CherryPit 1.2](https://huggingface.co/DRUPEs/CherryPit-1.2-2B)** by
@@ -56,14 +65,23 @@ Named after Vulcan, the Roman god of fire and the forge, the smith who crafts to
56
 
57
  ## Results
58
 
 
 
 
59
  ### AIME 2026
60
 
61
  **23 / 30 = 76.7%** with thinking on, up to 24K tokens, temperature 1.0, 1 sample, Q4_K_M.
62
  Per-problem results: [`aime2026_results.md`](aime2026_results.md).
63
 
 
 
 
64
  ### Selection league
65
 
66
- Scored on hard coding questions, plus some maths, agentic and reasoning tasks
 
 
 
67
  (0 / 0.5 / 1 per answer, Q4_K_M, temperature 1.0). Vulcan finished first. Plain CherryPit 1.2 is shown as the baseline.
68
 
69
  ![league results, part 1](images/league-1.png)
 
32
  - vulcan
33
  ---
34
 
35
+ For questions, contact ryan@filioai.com
36
+
37
+ Problems I've solved:
38
+
39
+ 1) Instant generation of a brand new GGUF file (takes seconds to minutes for larger models, done during the trials so there's no downtime.)
40
+ 2) Natural selection replicated with great success, in just a few rounds we see success using a specific technique.
41
+ 3) All done in quantised space without ever making a safetensors file. After, the GGUF can be reversed to safetensors.
42
+ 4) Tensor analysis techniques taking into account the architecture of the model and the shape of it.
43
+
44
  # Vulcan-1-Coder (GGUF)
45
 
46
  A 2.5B coding and reasoning model built on **[CherryPit 1.2](https://huggingface.co/DRUPEs/CherryPit-1.2-2B)** by
 
65
 
66
  ## Results
67
 
68
+ Notes: This model has not been fully tested or trained further after selection. It may have flaws or issues, for example one time it send a bunch of symbols.
69
+ But generally it scores high.
70
+
71
  ### AIME 2026
72
 
73
  **23 / 30 = 76.7%** with thinking on, up to 24K tokens, temperature 1.0, 1 sample, Q4_K_M.
74
  Per-problem results: [`aime2026_results.md`](aime2026_results.md).
75
 
76
+ Run 2:
77
+ **26 / 30 = 86.7%** with thinking on, up to 24k tokens, temperature 1.0, 1 sample, Q8, selectre
78
+
79
  ### Selection league
80
 
81
+ Scored on hard coding questions, plus some maths, agentic and reasoning tasks. No repeat questions and a large pool to avoid over-selecting for benchmarks.
82
+ NO BENCHMARK QUESTIONS, all generated by Claude including some questions that are extremely hard. The league questions varied between models and the questions
83
+ were not enough to find true representative results.
84
+
85
  (0 / 0.5 / 1 per answer, Q4_K_M, temperature 1.0). Vulcan finished first. Plain CherryPit 1.2 is shown as the baseline.
86
 
87
  ![league results, part 1](images/league-1.png)