Non-compositional Qwen3-8B L24 warmstart data, doc-level train/val split

README.md

@@ -0,0 +1,76 @@
+---
+license: odc-by
+configs:
+- config_name: av_sft
+  data_files:
+  - split: train
+    path: av_sft_train.parquet
+  - split: validation
+    path: av_sft_val.parquet
+- config_name: ar_sft
+  data_files:
+  - split: train
+    path: ar_sft_train.parquet
+  - split: validation
+    path: ar_sft_val.parquet
+tags:
+- interpretability
+- activations
+---
+
+# EasyNLA warmstart data (non-compositional, Qwen3-8B layer 24)
+
+Supervised warm-start data for [EasyNLA](https://github.com/asherps/EasyNLA) —
+train a **natural-language autoencoder** on Qwen3-8B: an activation-verbalizer
+(AV) that explains a layer-24 residual activation in natural language, and an
+activation-reconstructor (AR) that maps the explanation back to the activation.
+
+Each row carries one **raw** (unnormalized) layer-24 activation captured while
+Qwen3-8B read a FineFineWeb document (position ≥ 50 tokens), plus a
+Claude-written "gold" explanation of what the model was representing there.
+
+| config | split | rows | docs |
+|---|---|---|---|
+| `av_sft` | train | 363,961 | 36,734 |
+| `av_sft` | validation | 7,307 | 737 |
+| `ar_sft` | train | 363,988 | 36,734 |
+| `ar_sft` | validation | 7,405 | 747 |
+
+## Columns
+
+- `prompt` — chat messages for the AV (with the `<INJECT>` marker placeholder)
+  in `av_sft`; the AR critic prompt text in `ar_sft`
+- `response` (`av_sft` only) — the gold explanation the AV imitates
+- `activation_vector` — raw layer-24 residual, `list<float32>[4096]`
+- `doc_id`, `n_raw_tokens`, `activation_layer`,
+  `detokenized_text_truncated` (the source text up to the extraction position)
+
+Each parquet has a matching `<name>.parquet.nla_meta.yaml` **sidecar** — the
+contract EasyNLA trainers assert at startup (marker token IDs, prompt
+templates, `d_model`, scales). Keep them next to the parquets.
+
+## Splits
+
+Document-level and deterministic: a doc goes to **validation** iff
+`zlib.crc32(doc_id) % 1000 < 20` (`nla/val_split.py::is_val_doc(doc_id, 20)`,
+~2% of docs). No document's rows are split across train/validation, and the
+`av_sft` / `ar_sft` configs come from disjoint document sets by construction
+(stage-1 of the EasyNLA datagen pipeline).
+
+## Use with EasyNLA
+
+```bash
+huggingface-cli download asher577/easynla-warmstart-data --repo-type dataset \
+    --local-dir <data>
+
+python -m nla.train_sft --mode av --base-ckpt Qwen/Qwen3-8B \
+    --parquet <data>/av_sft_train.parquet --sidecar <data>/av_sft_train.parquet \
+    --heldout-parquet <data>/av_sft_val.parquet --save-dir <ckpts>/av
+python -m nla.train_sft --mode ar --base-ckpt Qwen/Qwen3-8B \
+    --parquet <data>/ar_sft_train.parquet --sidecar <data>/ar_sft_train.parquet \
+    --heldout-parquet <data>/ar_sft_val.parquet --save-dir <ckpts>/ar
+```
+
+Provenance: source text from FineFineWeb (ODC-By); explanations generated with
+Claude (claude-sonnet-4-6); activations from Qwen/Qwen3-8B. The `doc_id`
+strings are opaque keys from the generating pipeline.

ar_sft_train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af80520af3ddc6b0e30486d526f9f5a2188b3404699e535c144b2b392cd5fd4c
+size 2893128989

ar_sft_train.parquet.nla_meta.yaml

@@ -0,0 +1,77 @@
+dataset_id: ar_sft_Qwen3-8B_L24_6a22e324__ar_sft_raw__explained__shuf42
+stage: ar_sft
+row_count: 247358
+extraction:
+  base_model: Qwen/Qwen3-8B
+  d_model: 4096
+  layer_index: 24
+  norm: none
+  corpus: /workspace-vast/celeste/nla-data/finefineweb_100k.parquet
+  corpus_slice:
+    start: 0
+    length: 100000
+  positions_per_doc: 10
+kind: nla_dataset
+schema_version: 1
+keep_debug_metadata: true
+tokens:
+  injection_char: "\u320E"
+  injection_token_id: 149705
+  injection_left_neighbor_id: 29
+  injection_right_neighbor_id: 522
+  critic_suffix_ids:
+  - 1318
+  - 29
+  - 366
+  - 1708
+  - 29
+prompt_templates:
+  actor: 'You are a meticulous AI researcher conducting an important investigation
+    into activation vectors from a language model. Your overall task is to describe
+    the semantic content of that activation vector.
+
+
+    We will pass the vector enclosed in <concept> tags into your context. You must
+    then produce an explanation for the vector, enclosed within <explanation> tags.
+    The explanation consists of 2-3 text snippets describing that vector.
+
+
+    Here is the vector:
+
+
+    <concept>{injection_char}</concept>
+
+
+    Please provide an explanation.'
+  critic: 'Summary of the following text: <text>{explanation}</text> <summary>'
+api_summaries:
+  model: claude-sonnet-4-6
+  max_tokens: 300
+  temperature: 1.0
+  instruction_prompt: "A language model needs to predict what text comes next after\
+    \ a snippet which will be presented to you shortly. Identify the 2-3 most important\
+    \ features it would use for this prediction.\nFocus on what the language model\
+    \ must be \"thinking about\" at the point where the provided text ends. You should\
+    \ not need to reference the fact that the text is truncated/incomplete/a prefix:\
+    \ the language model is causal, so only sees the prefix to what it predicts and\
+    \ this is implicit.\nOrder features by what is most important for predicting the\
+    \ next tokens. Each feature should consist of a concise ~10-20 word description.\
+    \ Feel free to include specific textual examples inline.\n\nFeature types to consider\
+    \ (as inspiration, not a rigid checklist):\n- Syntactic/structural constraints:\
+    \ \"unclosed parenthesis requires matching close\"\n- Immediate semantic expectations:\
+    \ \"list promised three items but only two given\"\n- Stylistic/register patterns:\
+    \ \"formal academic tone maintained throughout\"\n- Narrative/argumentative momentum:\
+    \ \"thesis stated, supporting evidence now expected\"\n- Domain/genre signals:\
+    \ \"medical case history following SOAP format\"\n- Repetition/continuation patterns:\
+    \ \"same phrase structure repeating with variations\"\n\nThe final feature must\
+    \ describe the very end of the presented sequence: its role, what it's part of,\
+    \ and immediate constraints on what follows.\n\nFormat \u2014 IMPORTANT: keep\
+    \ to ~80-100 words total and ALWAYS close the tag:\n<analysis>\n[first feature\
+    \ \u2014 include specific examples when relevant]\n[second feature]\n[final feature:\
+    \ the last token, its role, immediate constraints]\n</analysis>\n\nText to analyze:\n\
+    \n<begin_text>{text}<end_text>"
+parent_datasets:
+- ar_sft_Qwen3-8B_L24_6a22e324__ar_sft_raw__explained
+created_at: '2026-05-16T19:57:10.471220+00:00'
+created_by: nla.datagen.stage_shuffle
+git_commit: 047eb8e

ar_sft_val.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c9d92622fbd51849e532d8be0db24ad20a15ac53f340e8fa6c32da2bf58e4816
+size 55102932

ar_sft_val.parquet.nla_meta.yaml

@@ -0,0 +1,77 @@
+dataset_id: ar_sft_Qwen3-8B_L24_6a22e324__ar_sft_raw__explained__shuf42
+stage: ar_sft
+row_count: 247358
+extraction:
+  base_model: Qwen/Qwen3-8B
+  d_model: 4096
+  layer_index: 24
+  norm: none
+  corpus: /workspace-vast/celeste/nla-data/finefineweb_100k.parquet
+  corpus_slice:
+    start: 0
+    length: 100000
+  positions_per_doc: 10
+kind: nla_dataset
+schema_version: 1
+keep_debug_metadata: true
+tokens:
+  injection_char: "\u320E"
+  injection_token_id: 149705
+  injection_left_neighbor_id: 29
+  injection_right_neighbor_id: 522
+  critic_suffix_ids:
+  - 1318
+  - 29
+  - 366
+  - 1708
+  - 29
+prompt_templates:
+  actor: 'You are a meticulous AI researcher conducting an important investigation
+    into activation vectors from a language model. Your overall task is to describe
+    the semantic content of that activation vector.
+
+
+    We will pass the vector enclosed in <concept> tags into your context. You must
+    then produce an explanation for the vector, enclosed within <explanation> tags.
+    The explanation consists of 2-3 text snippets describing that vector.
+
+
+    Here is the vector:
+
+
+    <concept>{injection_char}</concept>
+
+
+    Please provide an explanation.'
+  critic: 'Summary of the following text: <text>{explanation}</text> <summary>'
+api_summaries:
+  model: claude-sonnet-4-6
+  max_tokens: 300
+  temperature: 1.0
+  instruction_prompt: "A language model needs to predict what text comes next after\
+    \ a snippet which will be presented to you shortly. Identify the 2-3 most important\
+    \ features it would use for this prediction.\nFocus on what the language model\
+    \ must be \"thinking about\" at the point where the provided text ends. You should\
+    \ not need to reference the fact that the text is truncated/incomplete/a prefix:\
+    \ the language model is causal, so only sees the prefix to what it predicts and\
+    \ this is implicit.\nOrder features by what is most important for predicting the\
+    \ next tokens. Each feature should consist of a concise ~10-20 word description.\
+    \ Feel free to include specific textual examples inline.\n\nFeature types to consider\
+    \ (as inspiration, not a rigid checklist):\n- Syntactic/structural constraints:\
+    \ \"unclosed parenthesis requires matching close\"\n- Immediate semantic expectations:\
+    \ \"list promised three items but only two given\"\n- Stylistic/register patterns:\
+    \ \"formal academic tone maintained throughout\"\n- Narrative/argumentative momentum:\
+    \ \"thesis stated, supporting evidence now expected\"\n- Domain/genre signals:\
+    \ \"medical case history following SOAP format\"\n- Repetition/continuation patterns:\
+    \ \"same phrase structure repeating with variations\"\n\nThe final feature must\
+    \ describe the very end of the presented sequence: its role, what it's part of,\
+    \ and immediate constraints on what follows.\n\nFormat \u2014 IMPORTANT: keep\
+    \ to ~80-100 words total and ALWAYS close the tag:\n<analysis>\n[first feature\
+    \ \u2014 include specific examples when relevant]\n[second feature]\n[final feature:\
+    \ the last token, its role, immediate constraints]\n</analysis>\n\nText to analyze:\n\
+    \n<begin_text>{text}<end_text>"
+parent_datasets:
+- ar_sft_Qwen3-8B_L24_6a22e324__ar_sft_raw__explained
+created_at: '2026-05-16T19:57:10.471220+00:00'
+created_by: nla.datagen.stage_shuffle
+git_commit: 047eb8e

av_sft_train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac7011a41c38b23318e2df5b895ff3b49a3e66f70e84553d184052f36879d511
+size 2891503962

av_sft_train.parquet.nla_meta.yaml

@@ -0,0 +1,72 @@
+dataset_id: av_sft_Qwen3-8B_L24_6a22e324__av_sft_raw__explained__shuf42
+stage: av_sft
+row_count: 247261
+extraction:
+  base_model: Qwen/Qwen3-8B
+  d_model: 4096
+  layer_index: 24
+  norm: none
+  corpus: /workspace-vast/celeste/nla-data/finefineweb_100k.parquet
+  corpus_slice:
+    start: 0
+    length: 100000
+  positions_per_doc: 10
+kind: nla_dataset
+schema_version: 1
+keep_debug_metadata: true
+tokens:
+  injection_char: "\u320E"
+  injection_token_id: 149705
+  injection_left_neighbor_id: 29
+  injection_right_neighbor_id: 522
+  critic_suffix_ids: null
+prompt_templates:
+  actor: 'You are a meticulous AI researcher conducting an important investigation
+    into activation vectors from a language model. Your overall task is to describe
+    the semantic content of that activation vector.
+
+
+    We will pass the vector enclosed in <concept> tags into your context. You must
+    then produce an explanation for the vector, enclosed within <explanation> tags.
+    The explanation consists of 2-3 text snippets describing that vector.
+
+
+    Here is the vector:
+
+
+    <concept>{injection_char}</concept>
+
+
+    Please provide an explanation.'
+  critic: 'Summary of the following text: <text>{explanation}</text> <summary>'
+api_summaries:
+  model: claude-sonnet-4-6
+  max_tokens: 300
+  temperature: 1.0
+  instruction_prompt: "A language model needs to predict what text comes next after\
+    \ a snippet which will be presented to you shortly. Identify the 2-3 most important\
+    \ features it would use for this prediction.\nFocus on what the language model\
+    \ must be \"thinking about\" at the point where the provided text ends. You should\
+    \ not need to reference the fact that the text is truncated/incomplete/a prefix:\
+    \ the language model is causal, so only sees the prefix to what it predicts and\
+    \ this is implicit.\nOrder features by what is most important for predicting the\
+    \ next tokens. Each feature should consist of a concise ~10-20 word description.\
+    \ Feel free to include specific textual examples inline.\n\nFeature types to consider\
+    \ (as inspiration, not a rigid checklist):\n- Syntactic/structural constraints:\
+    \ \"unclosed parenthesis requires matching close\"\n- Immediate semantic expectations:\
+    \ \"list promised three items but only two given\"\n- Stylistic/register patterns:\
+    \ \"formal academic tone maintained throughout\"\n- Narrative/argumentative momentum:\
+    \ \"thesis stated, supporting evidence now expected\"\n- Domain/genre signals:\
+    \ \"medical case history following SOAP format\"\n- Repetition/continuation patterns:\
+    \ \"same phrase structure repeating with variations\"\n\nThe final feature must\
+    \ describe the very end of the presented sequence: its role, what it's part of,\
+    \ and immediate constraints on what follows.\n\nFormat \u2014 IMPORTANT: keep\
+    \ to ~80-100 words total and ALWAYS close the tag:\n<analysis>\n[first feature\
+    \ \u2014 include specific examples when relevant]\n[second feature]\n[final feature:\
+    \ the last token, its role, immediate constraints]\n</analysis>\n\nText to analyze:\n\
+    \n<begin_text>{text}<end_text>"
+parent_datasets:
+- av_sft_Qwen3-8B_L24_6a22e324__av_sft_raw__explained
+created_at: '2026-05-16T19:56:05.443944+00:00'
+created_by: nla.datagen.stage_shuffle
+git_commit: 047eb8e

av_sft_val.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f0f3deee370bec0873ac885d7b42182f3d21bdb158d237974183a29ba32eb5f
+size 53923447

av_sft_val.parquet.nla_meta.yaml

@@ -0,0 +1,72 @@
+dataset_id: av_sft_Qwen3-8B_L24_6a22e324__av_sft_raw__explained__shuf42
+stage: av_sft
+row_count: 247261
+extraction:
+  base_model: Qwen/Qwen3-8B
+  d_model: 4096
+  layer_index: 24
+  norm: none
+  corpus: /workspace-vast/celeste/nla-data/finefineweb_100k.parquet
+  corpus_slice:
+    start: 0
+    length: 100000
+  positions_per_doc: 10
+kind: nla_dataset
+schema_version: 1
+keep_debug_metadata: true
+tokens:
+  injection_char: "\u320E"
+  injection_token_id: 149705
+  injection_left_neighbor_id: 29
+  injection_right_neighbor_id: 522
+  critic_suffix_ids: null
+prompt_templates:
+  actor: 'You are a meticulous AI researcher conducting an important investigation
+    into activation vectors from a language model. Your overall task is to describe
+    the semantic content of that activation vector.
+
+
+    We will pass the vector enclosed in <concept> tags into your context. You must
+    then produce an explanation for the vector, enclosed within <explanation> tags.
+    The explanation consists of 2-3 text snippets describing that vector.
+
+
+    Here is the vector:
+
+
+    <concept>{injection_char}</concept>
+
+
+    Please provide an explanation.'
+  critic: 'Summary of the following text: <text>{explanation}</text> <summary>'
+api_summaries:
+  model: claude-sonnet-4-6
+  max_tokens: 300
+  temperature: 1.0
+  instruction_prompt: "A language model needs to predict what text comes next after\
+    \ a snippet which will be presented to you shortly. Identify the 2-3 most important\
+    \ features it would use for this prediction.\nFocus on what the language model\
+    \ must be \"thinking about\" at the point where the provided text ends. You should\
+    \ not need to reference the fact that the text is truncated/incomplete/a prefix:\
+    \ the language model is causal, so only sees the prefix to what it predicts and\
+    \ this is implicit.\nOrder features by what is most important for predicting the\
+    \ next tokens. Each feature should consist of a concise ~10-20 word description.\
+    \ Feel free to include specific textual examples inline.\n\nFeature types to consider\
+    \ (as inspiration, not a rigid checklist):\n- Syntactic/structural constraints:\
+    \ \"unclosed parenthesis requires matching close\"\n- Immediate semantic expectations:\
+    \ \"list promised three items but only two given\"\n- Stylistic/register patterns:\
+    \ \"formal academic tone maintained throughout\"\n- Narrative/argumentative momentum:\
+    \ \"thesis stated, supporting evidence now expected\"\n- Domain/genre signals:\
+    \ \"medical case history following SOAP format\"\n- Repetition/continuation patterns:\
+    \ \"same phrase structure repeating with variations\"\n\nThe final feature must\
+    \ describe the very end of the presented sequence: its role, what it's part of,\
+    \ and immediate constraints on what follows.\n\nFormat \u2014 IMPORTANT: keep\
+    \ to ~80-100 words total and ALWAYS close the tag:\n<analysis>\n[first feature\
+    \ \u2014 include specific examples when relevant]\n[second feature]\n[final feature:\
+    \ the last token, its role, immediate constraints]\n</analysis>\n\nText to analyze:\n\
+    \n<begin_text>{text}<end_text>"
+parent_datasets:
+- av_sft_Qwen3-8B_L24_6a22e324__av_sft_raw__explained
+created_at: '2026-05-16T19:56:05.443944+00:00'
+created_by: nla.datagen.stage_shuffle
+git_commit: 047eb8e