I guess for a 300B parameter or more and couple million users with the price of storage increasing as part of ramagedon this is also not viable...

I stopped paying and stayed in 2024.2, every once in a while I see if there is anything meaningful worth it there... nop

Sometimes I need to explain an issue, elaborate on the constraints, ask the model to research what's going on but I have to fight the model NON STOP to not start writing code... the solution is not to make the code pass where it's broken "fixing" all the unit test to now pass where they fail after the "fix"...

To provide the context so a good solution is done and taking into consideration EVERYTHING THAT needs to be done I need to "have a talk" with the model so I can present all the things to consider...

Today reading the output I got the two pieces below... first... this totally breaks the model for anything serious.. Why are you pushing the model to do something when I want it to do something else? I'M PAYING FOR IT,

But now in the middle of this RANT I notice the model mention that the "Auto pilot" mode is actually pressuring it to assume things and move on to complete as quick as possible...

I hope someone on Microsoft reads this... whenever you set up auto pilot.. you should put a message saying "the model main aim is to complete the task, and will make assumptions that facilitate that goal" which is a polite way of saying "auto pilot = model doesn't care the goal is to complete the task... either wrong or extremely wrong"

=============================================== ◐ The user is actively in conversation with me, walking through the problem step by step. They're frustrated that I keep trying to close out the task. They want to reason through this together. But the system keeps pushing me to complete. They haven't given me the next direction yet. The system autopilot is pushing me to act, but the user explicitly said "DO NOT MODIFY CODE" and wants to reason through this.

  I'm in a bind - the user wants a conversation, the system wants completion. 

The user is frustrated. They explicitly said not to modify code and to reason through this together. I searched to understand the scope of the problem - that's research, not implementation. But the system keeps telling me to implement and call task_complete. The user wants to have a design conversation.

Comments URL: https://news.ycombinator.com/item?id=48096339

Points: 2

# Comments: 0

But with this patch I saw 46t/s with qwen3.6 27B q8... this is insane, it's 250% faster than the original speed, there was no gpu I could upgrade to get that kind of boost, amazing!

The other thing, when anthropic turns on lazy claude... (I want to coin here the term Claudez for the version of claude that's lazy.. Claude zzZZzz = Claudez) that thing is terrible... you ask the model for something... and it's like... oh yes, that will probably depend on memory bandwith... do you want me to search that?...

YES... DO IT... FRICKING MACHINE..

Comments URL: https://news.ycombinator.com/item?id=47726788

Points: 4

# Comments: 1

The results are more spectacular...

The model pointed way better in gsm8k, but lost a bit on the other categories.

On the prior art: you're right that layer duplication has been explored before. What I think is new here is the systematic sweep toolkit + validation on standard benchmarks (lm-eval BBH, GSM8K, MBPP) showing exactly which 3 layers matter for which model. The Devstral logical deduction result (0.22→0.76) was a surprise to me.

If there are ComfyUI nodes that do this for image models, I'd love links, the "cognitive modes" finding (different duplication patterns that leads to different capability profiles from the same weights) might be even more interesting for diffusion models.

  # Run lm-evaluation-harness
  lm_eval --model local-chat-completions \
      --model_args model=test,base_url=http://localhost:8089/v1/chat/completions,num_concurrent=1,max_retries=3,tokenized_requests=False \
      --tasks gsm8k_cot,ifeval,mbpp,bbh_cot_fewshot_logical_deduction_five_objects,mbpp \
      --apply_chat_template --limit 50 \
      --output_path ./eval_results

I chat with the model to see if the thing was still working and seemed coherent to me, I didn't notice anything off.

I need to automate testing like that, where you pick the local maxima and then iterate over that picking layers to see if it's actually better, and then leave the thing running overnight

if this is validated enough it can eventually lead to ship some kind of "mix" architecture with layers executed to fit some "vibe?"

Devstral was the first one I tried and optimize for math/eq, but that din't result in any better model, then I added the reason part, and that resulted in "better" model

I used the devstral with the vibe.cli and it look sharp to me, thing didn't fail, I also used the chat to "vibe" check it and look ok to me.

The other thing is that I pick a particular circuit and that was "good" but I don't know if it was a local maxima, I think I ran just like 10 sets of the "fast test harness" and pick the config that gave the most score... once I have that I use that model and run it against the llm_eval limited to only 50 tests... again for sake of speed, I didn't want to wait a week to discover the config was bad

I'm using the following configuration --tasks gsm8k_cot,ifeval,mbpp,bbh_cot_fewshot_logical_deduction_five_objects,mbpp I did also try humaneval but something in the harness is missing and failed...

notice that I'm running 50 tests for each task, mostly because of time limitation as it takes like two hours to validate the run for the base model and the modified one.

I'll also try to publish the results of the small tests harness when I'm testing the multiple layers configurations, for reference this is phi-4-Q6_K.gguf, still running, I'm now giving more importance to the Reason factor, the reason factor comes from running a small subset of all the problems in the task config above

Initially I tried the approach of the highest math/eq but in resulted in models that were less capable overall with the exception of math, and math like in the original research is basically how good was the model at giving you the answer of a really though question, say the cubic root of some really large number... but that didn't translate to the model being better at other tasks...

  Config  | Lyr | Math   | EQ    | Reas   | Math Δ  | EQ Δ  | Reas Δ  | Comb Δ
  --------|-----|--------|-------|--------|---------|-------|---------|-------
  BASE    |   0 | 0.7405 | 94.49 | 94.12% |     --- |   --- |     --- |    ---
  (6,9)   |   3 | 0.7806 | 95.70 | 94.12% | +0.0401 | +1.21 |  +0.00% |  +1.21
  (9,12)  |   3 | 0.7247 | 95.04 | 94.12% | -0.0158 | +0.55 |  +0.00% |  +0.55
  (12,15) |   3 | 0.7258 | 94.14 | 88.24% | -0.0147 | -0.35 |  -5.88% |  -6.23
  (15,18) |   3 | 0.7493 | 95.74 | 88.24% | +0.0088 | +1.25 |  -5.88% |  -4.63
  (18,21) |   3 | 0.7204 | 93.40 | 94.12% | -0.0201 | -1.09 |  +0.00% |  -1.09
  (21,24) |   3 | 0.7107 | 92.97 | 88.24% | -0.0298 | -1.52 |  -5.88% |  -7.41
  (24,27) |   3 | 0.6487 | 95.27 | 88.24% | -0.0918 | +0.78 |  -5.88% |  -5.10
  (27,30) |   3 | 0.7180 | 94.65 | 88.24% | -0.0225 | +0.16 |  -5.88% |  -5.73
  (30,33) |   3 | 0.7139 | 94.02 | 94.12% | -0.0266 | -0.47 |  +0.00% |  -0.47
  (33,36) |   3 | 0.7104 | 94.53 | 94.12% | -0.0301 | +0.04 |  +0.00% |  +0.04
  (36,39) |   3 | 0.7017 | 94.69 | 94.12% | -0.0388 | +0.20 |  +0.00% |  +0.20
  (6,10)  |   4 | 0.8125 | 96.37 | 88.24% | +0.0720 | +1.88 |  -5.88% |  -4.01
  (9,13)  |   4 | 0.7598 | 95.08 | 94.12% | +0.0193 | +0.59 |  +0.00% |  +0.59
  (12,16) |   4 | 0.7482 | 93.71 | 88.24% | +0.0076 | -0.78 |  -5.88% |  -6.66
  (15,19) |   4 | 0.7617 | 95.16 | 82.35% | +0.0212 | +0.66 | -11.76% | -11.10
  (18,22) |   4 | 0.6902 | 92.27 | 88.24% | -0.0504 | -2.23 |  -5.88% |  -8.11
  (21,25) |   4 | 0.7288 | 94.10 | 88.24% | -0.0117 | -0.39 |  -5.88% |  -6.27
  (24,28) |   4 | 0.6823 | 94.57 | 88.24% | -0.0583 | +0.08 |  -5.88% |  -5.80
  (27,31) |   4 | 0.7224 | 94.41 | 82.35% | -0.0181 | -0.08 | -11.76% | -11.84
  (30,34) |   4 | 0.7070 | 94.73 | 94.12% | -0.0335 | +0.23 |  +0.00% |  +0.23
  (33,37) |   4 | 0.7009 | 94.38 |100.00% | -0.0396 | -0.12 |  +5.88% |  +5.77
  (36,40) |   4 | 0.7057 | 94.84 | 88.24% | -0.0348 | +0.35 |  -5.88% |  -5.53
  (6,11)  |   5 | 0.8168 | 95.62 |100.00% | +0.0762 | +1.13 |  +5.88% |  +7.02
  (9,14)  |   5 | 0.7245 | 95.23 | 88.24% | -0.0160 | +0.74 |  -5.88% |  -5.14
  (12,17) |   5 | 0.7825 | 94.88 | 88.24% | +0.0420 | +0.39 |  -5.88% |  -5.49
  (15,20) |   5 | 0.7832 | 95.86 | 88.24% | +0.0427 | +1.37 |  -5.88% |  -4.52
  (18,23) |   5 | 0.7208 | 92.42 | 88.24% | -0.0197 | -2.07 |  -5.88% |  -7.95
  (21,26) |   5 | 0.7055 | 92.89 | 88.24% | -0.0350 | -1.60 |  -5.88% |  -7.48
  (24,29) |   5 | 0.5825 | 95.04 | 94.12% | -0.1580 | +0.55 |  +0.00% |  +0.55
  (27,32) |   5 | 0.7088 | 94.18 | 88.24% | -0.0317 | -0.31 |  -5.88% |  -6.19
  (30,35) |   5 | 0.6787 | 94.69 | 88.24% | -0.0618 | +0.20 |  -5.88% |  -5.69
  (33,38) |   5 | 0.6650 | 94.96 | 88.24% | -0.0755 | +0.47 |  -5.88% |  -5.41
  (6,12)  |   6 | 0.7692 | 95.39 | 94.12% | +0.0287 | +0.90 |  +0.00% |  +0.90
  (9,15)  |   6 | 0.7405 | 94.65 | 94.12% | -0.0000 | +0.16 |  +0.00% |  +0.16
  (12,18) |   6 | 0.7582 | 94.57 | 88.24% | +0.0177 | +0.08 |  -5.88% |  -5.80
  (15,21) |   6 | 0.7828 | 93.52 | 88.24% | +0.0423 | -0.98 |  -5.88% |  -6.86
  (18,24) |   6 | 0.7308 | 92.93 | 94.12% | -0.0097 | -1.56 |  +0.00% |  -1.56
  (21,27) |   6 | 0.6791 | 92.54 | 82.35% | -0.0615 | -1.95 | -11.76% | -13.72

Transformers appear to have discrete "reasoning circuits" — contiguous blocks of 3-4 layers that act as indivisible cognitive units. Duplicate the right block and the model runs its reasoning pipeline twice. No weights change. No training. The model just thinks longer.

The results on standard benchmarks (lm-evaluation-harness, n=50):

Devstral-24B, layers 12-14 duplicated once: - BBH Logical Deduction: 0.22 → 0.76 - GSM8K (strict): 0.48 → 0.64 - MBPP (code gen): 0.72 → 0.78 - Nothing degraded

Qwen2.5-Coder-32B, layers 7-9 duplicated once: - Reasoning probe: 76% → 94%

The weird part: different duplication patterns create different cognitive "modes" from the same weights. Double-pass boosts math. Triple-pass boosts emotional reasoning. Interleaved doubling (13,13,14,14,15,15,16) creates a pure math specialist. Same model, same VRAM, different routing.

The circuit boundaries are sharp — shift by one layer and the effect disappears or inverts. Smaller models (24B) have tighter circuits (3 layers) than larger ones (Ng found 7 layers in 72B).

Tools to find circuits in any GGUF model and apply arbitrary layer routing are in the repo. The whole thing — sweep, discovery, validation — took one evening.

Happy to answer questions.

Comments URL: https://news.ycombinator.com/item?id=47431671

Points: 265

# Comments: 80

[0] https://www.steamdeck.com/en/tech [1] https://www.steamdeck.com/en/tech/deck