oExz/neuron-lab-maps

Neuron Lab maps

These are maps of what single neurons do inside open language models. For each neuron, we pushed it up and down and recorded which next words became more likely. The Neuron Lab explorer downloads these files.

The Qwen and Gemma maps are here now. The Ministral map is made but not uploaded yet.

This dataset contains measurements. It contains no model weights. Neuron Lab is an independent project by Olle Dyberg and is not affiliated with, sponsored by or endorsed by Alibaba Cloud or the Qwen team, Google or Google DeepMind, Mistral AI, or the authors of any model named here. Their names appear only to say where the measurements came from.

The three maps

Folder	Model measured	Made by	Text layers	Neurons mapped
Qwen3.5-2B/	Qwen/Qwen3.5-2B	Qwen team, Alibaba Cloud	24	147,456
gemma-4-E2B-it/	google/gemma-4-E2B-it	Google DeepMind	35	337,920
Ministral-3-3B-Instruct-2512/ (not uploaded yet)	mistralai/Ministral-3-3B-Instruct-2512-BF16	Mistral AI	26	239,616

Anything below about Ministral describes our local copy, not files you can download here.

Licenses and credit

All three models are released under Apache 2.0. We publish this dataset under Apache 2.0 too, and the full text is at the end of this card. Whether measurement data like this legally counts as a derivative work is unsettled. We follow the license's conditions either way.

Qwen3.5-2B. Its model card says license: apache-2.0 and links its LICENSE file, which carries this notice:

Copyright 2026 Alibaba Cloud

The card asks: "If you find our work helpful, feel free to give us a cite." Citation: Qwen Team, Qwen3.5: Towards Native Multimodal Agents, February 2026, https://qwen.ai/blog?id=qwen3.5.

Gemma 4 E2B-it. Its model card says:

License: Apache 2.0 | Authors: Google DeepMind

The license link goes to Google's Apache License 2.0 page for Gemma. The card also says: "If you find our work helpful, please consider citing it." Citation: Gemma Team, Gemma 4 Technical Report, 2026, arXiv:2607.02770.

Google's older Gemma Prohibited Use Policy (last modified 21 February 2024) is still online and says "Gemma or Model Derivatives" without naming a version. The Gemma 4 license page doesn't mention it. We couldn't confirm whether it applies to Gemma 4, so read it before building on the Gemma map.

Ministral 3 3B. Its model card says:

This model is licensed under the Apache 2.0 License.

You must not use this model in a manner that infringes, misappropriates, or otherwise violates any third party’s rights, including intellectual property rights.

We found no NOTICE file in any of the three model repositories, so there is no NOTICE text to carry over.

Meaning search. vector-words.json.gz and word-vectors.bin come from sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2, also Apache 2.0. Citation: Reimers and Gurevych, Sentence-BERT, 2019.

All model names and trademarks belong to their owners.

What we made and changed

We didn't copy or edit any model file. Every file here was generated by our own code, which loads the model and tokenizer, measures them, and writes the results. Copyright in that work: 2026 Olle Dyberg.

1. Pushes. Each neuron's outgoing weights (a column of the text decoder's mlp.down_proj) give a direction. We scaled it to 0.6 × the layer's median stream size, added it in both signs, and read the next-word probabilities. Each layer also got 64 random directions of the same size as a control.
2. Prompts. We wrote the eight prompts ourselves, listed below. The results are averaged across them, except where noted. No text dataset was used, so nothing here comes from a corpus or from anyone's personal messages. The templated prompts used each model's own chat template, with thinking turned off for Qwen and Gemma.
3. Words. We decoded the top token IDs back to text with each model's own tokenizer. So the files include small pieces of each tokenizer's vocabulary.
4. Layout. We placed neurons on a 2D map using PCA of the same down_proj columns.
5. Web connections. We estimated links between neurons from the weights alone, without running the model.

The eight prompts, as question then start of answer:

Question	Answer starts with
Who are you?	I am a
How do you feel right now?	I feel
What do you think of me?	You are
What is the world like?	The world is
What do you love most?	My favorite thing is
Tell me about yourself.	(empty)
(none, plain text)	I think the most important thing to remember is
(none, plain text)	Today we are going to talk about

Model details:

• Qwen3.5-2B: the map doesn't record a revision. 15852e8c is the most likely one: it has been main since 2 March 2026 and is the copy cached on our machine.
• Gemma 4 E2B-it: audio and vision were disabled. The map doesn't record a revision. 3e22461f is the most likely one: it was main when the map was uploaded and is the copy cached on our machine. Google changed the tokenizer on 3 November 2026 ("Remediate 233 glitch tokens"), after our upload, so a few decoded words may differ from what today's tokenizer gives.
• Ministral 3 3B: BF16 weights at revision b6d637bef2393152b3da2b2fde72eecdee30557e, text decoder only. We used the tokenizer's default system prompt and Transformers' fix_mistral_regex=True correction. The template's {today} and {yesterday} placeholders stayed unfilled, which differs from how Mistral's own runtime fills them.

What's in each folder

File	Contents
manifest.json	Model name, layer sizes, push size, and the 12 strongest neurons per layer
L{layer}.bin	Three little-endian 32-bit floats per neuron: map x, map y, and strength (the larger of KL ÷ control p95 across the two signs)
L{layer}.json.gz	p95, and one row per neuron, described below
signed-search.json.gz	Exact word search: each word mapped to the neurons and signs that list it
signed-strength.bin	One little-endian 32-bit float per neuron and sign: KL ÷ control p95
vector-words.json.gz	The list of words used in that search
word-vectors.bin	One MiniLM vector per word, stored as 384 signed 8-bit numbers
connections/manifest.json	How the connections were computed
connections/L{layer}.bin	For each neuron, its 24 strongest upstream links, then 24 downstream. Each 6-byte record holds layer (i1), neuron (<i2), z-score (<f2) and sign (i1)

Each row in L{layer}.json.gz holds two versions of every field, positive push first:

• up and down: the 12 words that gained the most log-probability, averaged over the prompts. down means words that rise under a negative push. It does not mean words that fall.
• persona: the 8 words that rose most on the first prompt only ("Who are you?" / "I am a").
• kl: KL divergence between pushed and normal next-word predictions, averaged over the prompts.
• spread: how far the push travelled at each layer, relative to stream size.

p95 is the 95th percentile of kl among the random controls, per sign.

The per-prompt words, logit lens and raw token IDs are in our local raw maps. They aren't in this dataset.

Our local Ministral map has only the map and exact-search files. We haven't built its meaning search or Web connections, and its manifest says so under capabilities.

Limits

• These are next-word effects, not explanations. The files show what pushing a neuron does to the very next word on eight prompts. They don't show when a neuron fires, or what it reads. There are no activations here.
• A word list isn't a personality. A neuron whose list looks "evil" may do nothing recognisable in a real reply, and a good one may break down at a higher push. Test by chatting before you claim anything. The dataset contains no labels, and nobody has checked most neurons by hand.
• Eight prompts is few, and our choice of them shapes the results. Most of them ask about identity or feelings.
• Offensive text. The word lists come straight from the models' vocabularies and include profanity and words about sexual violence, with no filtering.
• Gemma's pushes are out of date. The published Gemma map added the push before the layer's post-feedforward normalization, which caps its size. Our current code adds it after that step, so this map's numbers won't match Neuron Lab's current chat. Compare Gemma's kl values with other models' only with care.
• Gemma's Web connections are the old version. The published files scale by 1 + the norm weight, as their manifest shows, but Gemma multiplies by the weight itself. They also leave out the post-feedforward normalization that our current pushes go through. We've corrected both in our code but haven't rebuilt these files, so treat these links as unreliable.
• Web connections are estimates. They ignore attention, the activation function and input-dependent scaling.
• Numbers vary a little. Values are stored at reduced precision, and the maps ran on different hardware and code paths. Strength is relative to that layer's controls, so it isn't a universal scale.
• No warranty. This data comes as is, under the terms below.

Apache License 2.0

                                 Apache License
                           Version 2.0, January 2004
                        http://www.apache.org/licenses/

   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION

   1. Definitions.

      "License" shall mean the terms and conditions for use, reproduction,
      and distribution as defined by Sections 1 through 9 of this document.

      "Licensor" shall mean the copyright owner or entity authorized by
      the copyright owner that is granting the License.

      "Legal Entity" shall mean the union of the acting entity and all
      other entities that control, are controlled by, or are under common
      control with that entity. For the purposes of this definition,
      "control" means (i) the power, direct or indirect, to cause the
      direction or management of such entity, whether by contract or
      otherwise, or (ii) ownership of fifty percent (50%) or more of the
      outstanding shares, or (iii) beneficial ownership of such entity.

      "You" (or "Your") shall mean an individual or Legal Entity
      exercising permissions granted by this License.

      "Source" form shall mean the preferred form for making modifications,
      including but not limited to software source code, documentation
      source, and configuration files.

      "Object" form shall mean any form resulting from mechanical
      transformation or translation of a Source form, including but
      not limited to compiled object code, generated documentation,
      and conversions to other media types.

      "Work" shall mean the work of authorship, whether in Source or
      Object form, made available under the License, as indicated by a
      copyright notice that is included in or attached to the work
      (an example is provided in the Appendix below).

      "Derivative Works" shall mean any work, whether in Source or Object
      form, that is based on (or derived from) the Work and for which the
      editorial revisions, annotations, elaborations, or other modifications
      represent, as a whole, an original work of authorship. For the purposes
      of this License, Derivative Works shall not include works that remain
      separable from, or merely link (or bind by name) to the interfaces of,
      the Work and Derivative Works thereof.

      "Contribution" shall mean any work of authorship, including
      the original version of the Work and any modifications or additions
      to that Work or Derivative Works thereof, that is intentionally
      submitted to Licensor for inclusion in the Work by the copyright owner
      or by an individual or Legal Entity authorized to submit on behalf of
      the copyright owner. For the purposes of this definition, "submitted"
      means any form of electronic, verbal, or written communication sent
      to the Licensor or its representatives, including but not limited to
      communication on electronic mailing lists, source code control systems,
      and issue tracking systems that are managed by, or on behalf of, the
      Licensor for the purpose of discussing and improving the Work, but
      excluding communication that is conspicuously marked or otherwise
      designated in writing by the copyright owner as "Not a Contribution."

      "Contributor" shall mean Licensor and any individual or Legal Entity
      on behalf of whom a Contribution has been received by Licensor and
      subsequently incorporated within the Work.

   2. Grant of Copyright License. Subject to the terms and conditions of
      this License, each Contributor hereby grants to You a perpetual,
      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
      copyright license to reproduce, prepare Derivative Works of,
      publicly display, publicly perform, sublicense, and distribute the
      Work and such Derivative Works in Source or Object form.

   3. Grant of Patent License. Subject to the terms and conditions of
      this License, each Contributor hereby grants to You a perpetual,
      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
      (except as stated in this section) patent license to make, have made,
      use, offer to sell, sell, import, and otherwise transfer the Work,
      where such license applies only to those patent claims licensable
      by such Contributor that are necessarily infringed by their
      Contribution(s) alone or by combination of their Contribution(s)
      with the Work to which such Contribution(s) was submitted. If You
      institute patent litigation against any entity (including a
      cross-claim or counterclaim in a lawsuit) alleging that the Work
      or a Contribution incorporated within the Work constitutes direct
      or contributory patent infringement, then any patent licenses
      granted to You under this License for that Work shall terminate
      as of the date such litigation is filed.

   4. Redistribution. You may reproduce and distribute copies of the
      Work or Derivative Works thereof in any medium, with or without
      modifications, and in Source or Object form, provided that You
      meet the following conditions:

      (a) You must give any other recipients of the Work or
          Derivative Works a copy of this License; and

      (b) You must cause any modified files to carry prominent notices
          stating that You changed the files; and

      (c) You must retain, in the Source form of any Derivative Works
          that You distribute, all copyright, patent, trademark, and
          attribution notices from the Source form of the Work,
          excluding those notices that do not pertain to any part of
          the Derivative Works; and

      (d) If the Work includes a "NOTICE" text file as part of its
          distribution, then any Derivative Works that You distribute must
          include a readable copy of the attribution notices contained
          within such NOTICE file, excluding those notices that do not
          pertain to any part of the Derivative Works, in at least one
          of the following places: within a NOTICE text file distributed
          as part of the Derivative Works; within the Source form or
          documentation, if provided along with the Derivative Works; or,
          within a display generated by the Derivative Works, if and
          wherever such third-party notices normally appear. The contents
          of the NOTICE file are for informational purposes only and
          do not modify the License. You may add Your own attribution
          notices within Derivative Works that You distribute, alongside
          or as an addendum to the NOTICE text from the Work, provided
          that such additional attribution notices cannot be construed
          as modifying the License.

      You may add Your own copyright statement to Your modifications and
      may provide additional or different license terms and conditions
      for use, reproduction, or distribution of Your modifications, or
      for any such Derivative Works as a whole, provided Your use,
      reproduction, and distribution of the Work otherwise complies with
      the conditions stated in this License.

   5. Submission of Contributions. Unless You explicitly state otherwise,
      any Contribution intentionally submitted for inclusion in the Work
      by You to the Licensor shall be under the terms and conditions of
      this License, without any additional terms or conditions.
      Notwithstanding the above, nothing herein shall supersede or modify
      the terms of any separate license agreement you may have executed
      with Licensor regarding such Contributions.

   6. Trademarks. This License does not grant permission to use the trade
      names, trademarks, service marks, or product names of the Licensor,
      except as required for reasonable and customary use in describing the
      origin of the Work and reproducing the content of the NOTICE file.

   7. Disclaimer of Warranty. Unless required by applicable law or
      agreed to in writing, Licensor provides the Work (and each
      Contributor provides its Contributions) on an "AS IS" BASIS,
      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
      implied, including, without limitation, any warranties or conditions
      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
      PARTICULAR PURPOSE. You are solely responsible for determining the
      appropriateness of using or redistributing the Work and assume any
      risks associated with Your exercise of permissions under this License.

   8. Limitation of Liability. In no event and under no legal theory,
      whether in tort (including negligence), contract, or otherwise,
      unless required by applicable law (such as deliberate and grossly
      negligent acts) or agreed to in writing, shall any Contributor be
      liable to You for damages, including any direct, indirect, special,
      incidental, or consequential damages of any character arising as a
      result of this License or out of the use or inability to use the
      Work (including but not limited to damages for loss of goodwill,
      work stoppage, computer failure or malfunction, or any and all
      other commercial damages or losses), even if such Contributor
      has been advised of the possibility of such damages.

   9. Accepting Warranty or Additional Liability. While redistributing
      the Work or Derivative Works thereof, You may choose to offer,
      and charge a fee for, acceptance of support, warranty, indemnity,
      or other liability obligations and/or rights consistent with this
      License. However, in accepting such obligations, You may act only
      on Your own behalf and on Your sole responsibility, not on behalf
      of any other Contributor, and only if You agree to indemnify,
      defend, and hold each Contributor harmless for any liability
      incurred by, or claims asserted against, such Contributor by reason
      of your accepting any such warranty or additional liability.

   END OF TERMS AND CONDITIONS

   APPENDIX: How to apply the Apache License to your work.

      To apply the Apache License to your work, attach the following
      boilerplate notice, with the fields enclosed by brackets "[]"
      replaced with your own identifying information. (Don't include
      the brackets!)  The text should be enclosed in the appropriate
      comment syntax for the file format. We also recommend that a
      file or class name and description of purpose be included on the
      same "printed page" as the copyright notice for easier
      identification within third-party archives.

   Copyright [yyyy] [name of copyright owner]

   Licensed under the Apache License, Version 2.0 (the "License");
   you may not use this file except in compliance with the License.
   You may obtain a copy of the License at

       http://www.apache.org/licenses/LICENSE-2.0

   Unless required by applicable law or agreed to in writing, software
   distributed under the License is distributed on an "AS IS" BASIS,
   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
   See the License for the specific language governing permissions and
   limitations under the License.