Add sesion 32 paper-2 strengthening records (+21, total 79)

- Pythia-70M ν trajectory (9 ckpts × 2 corpora = 18 records)
honest null: within-model trajectory does not converge to -1/(2π)
- Yi-9B random_tokens (n=30) — 9B class gap filled
- Pythia-31m high-n robustness (n=60 mongo + n=60 random)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

README.md

@@ -49,11 +49,14 @@ where θ is the RoPE base frequency and T is the evaluation context length.
 
 ## Coverage
 
-- **32 models** across 12 families (Pythia, Qwen, Llama, Mistral, Gemma, Phi, OLMo, OLMoE, DeepSeek, StarCoder2, CodeLlama, GPT-J, SmolLM2, Falcon)
-- **58 records** total
+- **33 models** across 13 families (Pythia, Qwen, Llama, Mistral, Gemma, Phi, OLMo, OLMoE, DeepSeek, StarCoder2, CodeLlama, GPT-J, SmolLM2, Falcon, Yi)
+- **79 records** total
 - **2 corpora**: real text (`real_text`, MongoDB English episodes) + random tokens (`random_tokens`)
 - **2 precisions**: 4-bit NF4 (BitsAndBytes) + bfloat16
 - **Includes random-init controls** (E2 falsifier on Pythia 70M/410M/1B with random Gaussian init, no pretraining) — establishes that the slope ν = ∂γ/∂log₁₀(P) ≈ −1/(2π) is genuinely a *training imprint*, not architecture artifact.
+- **Pythia-70M training trajectory** (9 checkpoints × 2 corpora = 18 records, sesion 32) — within-model γ across `step1000` → `step143000`. **Honest null result**: trajectory does NOT converge to ν = −1/(2π); imprint constant emerges across-models, not within-model.
+- **Pythia-31m high-n robustness** (n=60 prompts × 2 corpora = 2 records) — tightens CI on smallest pythia anchor.
+- **Yi-9B random_tokens** (n=30) — fills 9B class gap in family panel.
 
 ## Schema
 
@@ -128,7 +131,7 @@ This release closes that gap.
 
 - **Raw attention tensors** (TB-scale, redundant with model weights)
 - **Per-layer per-head γ-fields** (separate dataset planned)
-- **Training-trajectory γ over checkpoints** (separate, from Pythia ckpts)
+- **Training-trajectory γ over checkpoints** (Pythia-70M trajectory now INCLUDED as of sesion 32; broader panel still planned)
 - **Downstream task scores** (use RULER, LongBench-v2, HELM separately)
 
 ## License
@@ -186,6 +189,10 @@ GitHub repo — refutations are welcome.
 - 2026-04-29: Initial release (58 records, 32 models, 2 corpora, 2 precisions)
 - 2026-04-30: Added `analysis/games/game_O_results.json` + `game_P_results.json`
   (hyperscaling identities + recursive derivations)
+- 2026-05-01: ★ **Sesion 32 paper 2 strengthening** — +21 records (79 total):
+  - Pythia-70M ν trajectory (9 ckpts × 2 corpora = 18) — within-model null documented
+  - Yi-9B random_tokens (1) — 9B class gap filled
+  - Pythia-31m high-n robustness (2, n=60 each) — tightened CI on smallest anchor
 - 2026-05-01: ★ **TAF v0.5 machine-verified consistency** — 15 algebraic identities
   of TAF critical exponents formally proven via dual-tool approach:
   - **Sage Groebner basis** (algebraic decision in PolynomialRing(ℚ))

summary.csv

@@ -57,3 +57,24 @@ Qwen/Qwen2.5-Coder-7B,qwen-coder,real_text,4-bit-NF4,0.9454,0.9547084761104221,7
 EleutherAI/pythia-1b,pythia,random_tokens,fp16,0.6818,,1000,10000,2048,False,False
 EleutherAI/pythia-410m,pythia,random_tokens,fp16,0.6845,,410,10000,2048,False,False
 EleutherAI/pythia-70m,pythia,random_tokens,fp16,0.6846,,70,10000,2048,False,False
+EleutherAI/pythia-31m,pythia,real_text,auto,1.0110361167221762,0.99999561666838,31,10000,2048,False,False
+EleutherAI/pythia-31m,pythia,random_tokens,auto,1.5450115522836714,0.99999561666838,31,10000,2048,False,False
+01-ai/Yi-9B,yi,random_tokens,auto,1.0608729832204116,,9000,5000000,4096,False,False
+EleutherAI/pythia-70m-deduped@step1000,pythia,real_text,auto,1.1292477525397862,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step1000,pythia,random_tokens,auto,1.0790853925598143,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step2000,pythia,real_text,auto,1.2389542356917203,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step2000,pythia,random_tokens,auto,1.1287893216072091,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step5000,pythia,real_text,auto,1.259578109732732,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step5000,pythia,random_tokens,auto,1.3075968596454313,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step10000,pythia,real_text,auto,1.2494402035559864,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step10000,pythia,random_tokens,auto,1.2512259968411465,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step20000,pythia,real_text,auto,1.24348361085288,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step20000,pythia,random_tokens,auto,1.2448392596274824,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step40000,pythia,real_text,auto,1.3158082124956427,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step40000,pythia,random_tokens,auto,1.2107643823126109,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step65000,pythia,real_text,auto,1.3110320441749503,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step65000,pythia,random_tokens,auto,1.2584672405293242,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step100000,pythia,real_text,auto,1.32261751836488,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step100000,pythia,random_tokens,auto,1.2204931999100648,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step143000,pythia,real_text,auto,1.2493746664709637,,70,10000,2048,False,False
+EleutherAI/pythia-70m-deduped@step143000,pythia,random_tokens,auto,1.2023289680609859,,70,10000,2048,False,False

taf-attention-decay.jsonl

@@ -56,3 +56,24 @@
 {"model_id": "EleutherAI/pythia-1b", "revision": "random-init-seed42", "arch": {"d_model": 2048, "n_heads": 8, "n_layers": 16, "d_head": 256, "n_kv_heads": 8, "n_params_M": 1000, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 0.6818490427805074, "gamma_ci95_lo": 0.5604049003213023, "gamma_ci95_hi": 0.8032931852397125, "method": "pade_d_alias_T", "fit": {"R2": 0.9527921778371977}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 100, "precision": "fp16"}, "predictions": {"note": "random-init falsifier (E2): predicts ν=0 if F1 mechanism is training-imprint"}, "decision": "F1 CONFIRMED (training-imprint mechanism)", "provenance": {"taf_version": "0.4", "experiment": "E2_imprint_init_falsifier", "source_file": "EleutherAI--pythia-1b_init.json", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0"}}
 {"model_id": "EleutherAI/pythia-410m", "revision": "random-init-seed42", "arch": {"d_model": 1024, "n_heads": 16, "n_layers": 24, "d_head": 64, "n_kv_heads": 16, "n_params_M": 410, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 0.6845471503266192, "gamma_ci95_lo": 0.5625861150445415, "gamma_ci95_hi": 0.8065081856086969, "method": "pade_d_alias_T", "fit": {"R2": 0.9527653674927113}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 100, "precision": "fp16"}, "predictions": {"note": "random-init falsifier (E2): predicts ν=0 if F1 mechanism is training-imprint"}, "decision": "F1 CONFIRMED (training-imprint mechanism)", "provenance": {"taf_version": "0.4", "experiment": "E2_imprint_init_falsifier", "source_file": "EleutherAI--pythia-410m_init.json", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0"}}
 {"model_id": "EleutherAI/pythia-70m", "revision": "random-init-seed42", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 0.6845950557028538, "gamma_ci95_lo": 0.5620692021162929, "gamma_ci95_hi": 0.8071209092894148, "method": "pade_d_alias_T", "fit": {"R2": 0.9523541015048161}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 100, "precision": "fp16"}, "predictions": {"note": "random-init falsifier (E2): predicts ν=0 if F1 mechanism is training-imprint"}, "decision": "F1 CONFIRMED (training-imprint mechanism)", "provenance": {"taf_version": "0.4", "experiment": "E2_imprint_init_falsifier", "source_file": "EleutherAI--pythia-70m_init.json", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0"}}
+{"model_id": "EleutherAI/pythia-31m", "revision": "main", "arch": {"d_model": 256, "n_heads": 4, "n_layers": 6, "d_head": 64, "n_kv_heads": 4, "n_params_M": 31, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.0110361167221762, "gamma_ci95_lo": 0.7204541110410775, "gamma_ci95_hi": 1.392663572169019, "method": "pade_d_alias_T", "fit": {"log_A": -1.7844995148362943, "R2": 0.879471, "n_points": 7, "delta_R2_power_minus_exp": 0.3602}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 60, "seeds": [42, 123, 7], "distances": [10, 20, 30, 50, 100, 200, 500, 1000, 2000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": 0.99999561666838, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "CONFIRMED: γ law holds (γ×ln(θ) = C)", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-31m_mongo.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-31m", "revision": "main", "arch": {"d_model": 256, "n_heads": 4, "n_layers": 6, "d_head": 64, "n_kv_heads": 4, "n_params_M": 31, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.5450115522836714, "gamma_ci95_lo": 1.151215335615751, "gamma_ci95_hi": 1.7644064236370867, "method": "pade_d_alias_T", "fit": {"log_A": 0.2663580390060624, "R2": 0.966846, "n_points": 7, "delta_R2_power_minus_exp": 0.0973}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 60, "seeds": [42, 123, 7], "distances": [10, 20, 30, 50, 100, 200, 500, 1000, 2000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": 0.99999561666838, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "ANOMALY: long-context training effect", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-31m_random.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "01-ai/Yi-9B", "revision": "main", "arch": {"d_model": 4096, "n_heads": 32, "n_layers": 48, "d_head": 128, "n_kv_heads": 4, "n_params_M": 9000, "rope_theta": 5000000, "T_train": 4096, "family": "yi", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.0608729832204116, "gamma_ci95_lo": 0.9164854642853202, "gamma_ci95_hi": 1.1751419131536853, "method": "pade_d_alias_T", "fit": {"log_A": -1.680130627799741, "R2": 0.988358, "n_points": 7, "delta_R2_power_minus_exp": 0.1638}, "T_eval": 4096, "corpus": "random_tokens", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 20, 30, 50, 100, 200, 500, 1000, 2000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "01-ai--Yi-9B_random.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step1000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.1292477525397862, "gamma_ci95_lo": 1.0315555906714828, "gamma_ci95_hi": 1.216214657750404, "method": "pade_d_alias_T", "fit": {"log_A": -0.6033750860926805, "R2": 0.997384, "n_points": 5, "delta_R2_power_minus_exp": 0.122}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_mongo_step1000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step1000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.0790853925598143, "gamma_ci95_lo": 1.017622968551251, "gamma_ci95_hi": 1.1097164598481626, "method": "pade_d_alias_T", "fit": {"log_A": -0.9672073904161836, "R2": 0.99908, "n_points": 5, "delta_R2_power_minus_exp": 0.1391}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_random_step1000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step2000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.2389542356917203, "gamma_ci95_lo": 1.0776344210678837, "gamma_ci95_hi": 1.3728486160782174, "method": "pade_d_alias_T", "fit": {"log_A": -0.3277183304755686, "R2": 0.993998, "n_points": 5, "delta_R2_power_minus_exp": 0.1126}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_mongo_step2000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step2000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.1287893216072091, "gamma_ci95_lo": 0.9179248583247418, "gamma_ci95_hi": 1.3789563868353336, "method": "pade_d_alias_T", "fit": {"log_A": -0.7034269751264387, "R2": 0.990493, "n_points": 5, "delta_R2_power_minus_exp": 0.0965}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_random_step2000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step5000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.259578109732732, "gamma_ci95_lo": 1.065451291650048, "gamma_ci95_hi": 1.425157147512835, "method": "pade_d_alias_T", "fit": {"log_A": -0.5097315436137113, "R2": 0.989525, "n_points": 5, "delta_R2_power_minus_exp": 0.1012}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_mongo_step5000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step5000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.3075968596454313, "gamma_ci95_lo": 1.08364183240963, "gamma_ci95_hi": 1.48036480213923, "method": "pade_d_alias_T", "fit": {"log_A": -0.11233545702411364, "R2": 0.988228, "n_points": 5, "delta_R2_power_minus_exp": 0.1023}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_random_step5000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step10000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.2494402035559864, "gamma_ci95_lo": 1.0870529096363453, "gamma_ci95_hi": 1.334282860775478, "method": "pade_d_alias_T", "fit": {"log_A": -0.7144769697219604, "R2": 0.994174, "n_points": 5, "delta_R2_power_minus_exp": 0.1243}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_mongo_step10000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step10000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.2512259968411465, "gamma_ci95_lo": 1.2173504871215475, "gamma_ci95_hi": 1.2761221205287583, "method": "pade_d_alias_T", "fit": {"log_A": -0.4228116863668217, "R2": 0.999617, "n_points": 5, "delta_R2_power_minus_exp": 0.141}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_random_step10000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step20000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.24348361085288, "gamma_ci95_lo": 1.025314490526864, "gamma_ci95_hi": 1.3802858861513394, "method": "pade_d_alias_T", "fit": {"log_A": -0.7971920795807635, "R2": 0.990125, "n_points": 5, "delta_R2_power_minus_exp": 0.1141}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_mongo_step20000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step20000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.2448392596274824, "gamma_ci95_lo": 1.1754331028426652, "gamma_ci95_hi": 1.3048233697356975, "method": "pade_d_alias_T", "fit": {"log_A": -0.5192663551206491, "R2": 0.996808, "n_points": 5, "delta_R2_power_minus_exp": 0.1264}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_random_step20000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step40000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.3158082124956427, "gamma_ci95_lo": 1.1294442761813686, "gamma_ci95_hi": 1.4329043163532948, "method": "pade_d_alias_T", "fit": {"log_A": -0.5386898247318178, "R2": 0.993845, "n_points": 5, "delta_R2_power_minus_exp": 0.1211}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_mongo_step40000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step40000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.2107643823126109, "gamma_ci95_lo": 1.0637363993275404, "gamma_ci95_hi": 1.3940139407758467, "method": "pade_d_alias_T", "fit": {"log_A": -0.7040201696142183, "R2": 0.993318, "n_points": 5, "delta_R2_power_minus_exp": 0.1155}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_random_step40000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step65000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.3110320441749503, "gamma_ci95_lo": 1.0407698744775105, "gamma_ci95_hi": 1.569371030798524, "method": "pade_d_alias_T", "fit": {"log_A": -0.6184523218509428, "R2": 0.978259, "n_points": 5, "delta_R2_power_minus_exp": 0.1439}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_mongo_step65000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step65000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.2584672405293242, "gamma_ci95_lo": 1.2368161267690767, "gamma_ci95_hi": 1.2753542562922369, "method": "pade_d_alias_T", "fit": {"log_A": -0.6381258873027861, "R2": 0.999895, "n_points": 5, "delta_R2_power_minus_exp": 0.1481}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_random_step65000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step100000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.32261751836488, "gamma_ci95_lo": 1.0829643214589222, "gamma_ci95_hi": 1.5030706982821522, "method": "pade_d_alias_T", "fit": {"log_A": -0.406669859016411, "R2": 0.983281, "n_points": 5, "delta_R2_power_minus_exp": 0.131}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_mongo_step100000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step100000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.2204931999100648, "gamma_ci95_lo": 1.0522202974778878, "gamma_ci95_hi": 1.3232134560235709, "method": "pade_d_alias_T", "fit": {"log_A": -0.8927465051706067, "R2": 0.992892, "n_points": 5, "delta_R2_power_minus_exp": 0.1325}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_random_step100000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step143000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.2493746664709637, "gamma_ci95_lo": 0.9036724881431525, "gamma_ci95_hi": 1.5802558154712143, "method": "pade_d_alias_T", "fit": {"log_A": -0.7806191250583789, "R2": 0.971774, "n_points": 5, "delta_R2_power_minus_exp": 0.1713}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_mongo_step143000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step143000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.2023289680609859, "gamma_ci95_lo": 0.9651031952002092, "gamma_ci95_hi": 1.4205245025802689, "method": "pade_d_alias_T", "fit": {"log_A": -0.9227782529370149, "R2": 0.986676, "n_points": 5, "delta_R2_power_minus_exp": 0.0937}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_random_step143000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}