Instructions to use pradhyut/evidencelock-state-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use pradhyut/evidencelock-state-lora with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("CohereLabs/tiny-aya-fire") model = PeftModel.from_pretrained(base_model, "pradhyut/evidencelock-state-lora") - Notebooks
- Google Colab
- Kaggle
Publish verified EVIDENCELOCK research adapters and provenance
Browse files- LICENSE.md +16 -0
- README.md +50 -0
- SHA256SUMS +7 -0
- adapter_config.json +46 -0
- adapter_model.safetensors +3 -0
- provenance-manifest.json +51 -0
- provenance/config-normalization.json +39 -0
- provenance/original_adapter_config.json +46 -0
LICENSE.md
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Adapter attribution and license notice
|
| 2 |
+
|
| 3 |
+
Base creator: Cohere Labs / Cohere. Base model: https://huggingface.co/CohereLabs/tiny-aya-fire
|
| 4 |
+
Base revision: 73500c80fcb6c8bea3e830c21d5d2920249a0e1a
|
| 5 |
+
Adaptations: EVIDENCELOCK contributors, 2026.
|
| 6 |
+
|
| 7 |
+
Adapter deltas are provided under Creative Commons Attribution-NonCommercial 4.0
|
| 8 |
+
with Cohere's acceptable-use addendum: https://cohere.com/cohere-labs-cc-by-nc-license
|
| 9 |
+
Cohere Labs Acceptable Use Policy: https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy
|
| 10 |
+
|
| 11 |
+
Retain upstream attribution, license/disclaimer links and modification notices.
|
| 12 |
+
The LoRA adaptation and metadata normalization are described in the model card
|
| 13 |
+
and provenance files. Attribution does not imply upstream endorsement.
|
| 14 |
+
Base weights and tokenizer assets are excluded and must be obtained separately.
|
| 15 |
+
This notice summarizes and links the applicable terms; it does not replace them.
|
| 16 |
+
Original EVIDENCELOCK source code is licensed separately under MIT.
|
README.md
ADDED
|
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-4.0
|
| 3 |
+
library_name: peft
|
| 4 |
+
pipeline_tag: text-generation
|
| 5 |
+
base_model: CohereLabs/tiny-aya-fire
|
| 6 |
+
base_model_relation: adapter
|
| 7 |
+
language:
|
| 8 |
+
- en
|
| 9 |
+
tags:
|
| 10 |
+
- peft
|
| 11 |
+
- lora
|
| 12 |
+
- cohere2
|
| 13 |
+
- evidencelock
|
| 14 |
+
- research
|
| 15 |
+
datasets:
|
| 16 |
+
- pradhyut/evidencelock-benchmark
|
| 17 |
+
---
|
| 18 |
+
# EVIDENCELOCK — state-only LoRA
|
| 19 |
+
|
| 20 |
+
This research adapter studies four evidence states: **SUPPORTED**, **MISSING_EVIDENCE**, **CONFLICT** and **OUT_OF_SCOPE**. It is a state classifier evaluated through next-token option scores; complete JSON response drafting and citation generation are a separate task.
|
| 21 |
+
|
| 22 |
+
## Training and decision protocol
|
| 23 |
+
|
| 24 |
+
The adapter completed 864 genuine optimizer updates using 864 curated exposures from 99 original synthetic training cases and 396 distinct case/option-rotation prompts. LoRA rank is 4, alpha 8, dropout 0, with q_proj and v_proj targets. The frozen state protocol uses four option rotations, scores the next-token options, takes a strict-majority decision and abstains without a strict majority. The source preserves the exact observable-ledger encoding and rotation protocol in `gpu/classification_protocol.py`, `gpu/compact_ledger.py` and the frozen scoring driver.
|
| 25 |
+
|
| 26 |
+
## Measured frozen English results
|
| 27 |
+
|
| 28 |
+
| Comparator | Correct state decisions | Abstentions | Conflict recall |
|
| 29 |
+
|---|---:|---:|---:|
|
| 30 |
+
| Base model with the same state protocol | 39/99 | 29/99 | See preserved per-case records |
|
| 31 |
+
| State-only LoRA | 77/99 | 1/99 | 9/9 |
|
| 32 |
+
|
| 33 |
+
The adapter correctly identified 20/27 missing-evidence cases, but called six missing-evidence cases SUPPORTED and abstained on one. It achieved 39/54 supported and 9/9 out-of-scope decisions. These measurements concern 99 synthetic held-out English scenarios, not broad task or language generalization. No held-out result was used for further tuning. State accuracy does not establish drafting, citation or deployment reliability.
|
| 34 |
+
|
| 35 |
+
The evidence benchmark uses coarse family/template-lineage separation. Development validation was reused during earlier experiments. Original factual-question versus draft-policy ambiguity is disclosed in the research records. The small translation-review subset was owner-approved with model assistance; independent native-speaker validation is not established. No multilingual classification result is asserted for this adapter.
|
| 36 |
+
|
| 37 |
+
## Attribution and artifact terms
|
| 38 |
+
|
| 39 |
+
Upstream model creator: **Cohere Labs / Cohere**. Base: [CohereLabs/tiny-aya-fire](https://huggingface.co/CohereLabs/tiny-aya-fire), revision `73500c80fcb6c8bea3e830c21d5d2920249a0e1a`.
|
| 40 |
+
EVIDENCELOCK contributors produced the LoRA adaptations. The original adapter weight bytes are preserved; only the public configuration's base identifier and revision metadata were normalized. Original configurations and exact before/after changes accompany each artifact.
|
| 41 |
+
|
| 42 |
+
These adapter deltas are distributed for noncommercial research under [CC-BY-NC 4.0 with Cohere's acceptable-use addendum](https://cohere.com/cohere-labs-cc-by-nc-license) and the [Cohere Labs Acceptable Use Policy](https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy). Retain attribution and modification notices. Attribution does not imply endorsement. Base weights and tokenizer assets are not redistributed; obtain the base separately under its access conditions. Original EVIDENCELOCK source has a separate MIT license.
|
| 43 |
+
|
| 44 |
+
## Reproduction and records
|
| 45 |
+
|
| 46 |
+
- [Source and GPU protocols](https://github.com/Pradhyut21/EVIDENCELOCK)
|
| 47 |
+
- [Synthetic benchmark and curated splits](https://huggingface.co/datasets/pradhyut/evidencelock-benchmark)
|
| 48 |
+
- [Raw outputs, native toolkit reports and execution records](https://huggingface.co/datasets/pradhyut/evidencelock-research-records)
|
| 49 |
+
|
| 50 |
+
The final Linux training and evaluation stack is Python 3.12.16, torch 2.11.0+cu130, Transformers 5.22.1, PEFT 0.21.2, TRL 1.7.1, Accelerate 1.15.0, bitsandbytes 0.50.2, CuratorKIT 0.1.22, AlignTune 0.1.28 and AuditKIT 1.1.3. Exact dependency locks and protocols are in the source repository; historical environments are recorded in their original reports. Frozen final inference uses FP16 compute, NF4 double quantization and the pinned model revision. Successful loading alone does not reproduce the measured scoring protocol.
|
SHA256SUMS
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
89eec08b76d40c589d69e290ef5069e724e4b6921bb41c9551bcff70bcf647ce adapter_config.json
|
| 2 |
+
cad9f72b601182473a88633ad5e68d703eb5f93a21360639276be7871a4b0fba adapter_model.safetensors
|
| 3 |
+
533b3b555e085f6212d8c3bac30cd71144137e6fb9f72ad70a5eca0abe0f6ff1 LICENSE.md
|
| 4 |
+
9888780414992c2d87b99659dabd32351779a9bdb3095acff7ccb24e10b5a47d provenance/config-normalization.json
|
| 5 |
+
44c9f947910559bfb3c56557badf4b0d393908c0c9d89a78beadf35e575b8333 provenance/original_adapter_config.json
|
| 6 |
+
21b5653a88521b4588ecc2e53fb5859f9dc8eb17b77bf38292b7b759bbd932af provenance-manifest.json
|
| 7 |
+
bf0e0c8794694bf90e7f22d1e85c2f769ff15a8578832b5e0a70171573a9c608 README.md
|
adapter_config.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"alora_invocation_tokens": null,
|
| 3 |
+
"alpha_pattern": {},
|
| 4 |
+
"arrow_config": null,
|
| 5 |
+
"auto_mapping": null,
|
| 6 |
+
"base_model_name_or_path": "CohereLabs/tiny-aya-fire",
|
| 7 |
+
"bias": "none",
|
| 8 |
+
"corda_config": null,
|
| 9 |
+
"ensure_weight_tying": false,
|
| 10 |
+
"eva_config": null,
|
| 11 |
+
"exclude_modules": null,
|
| 12 |
+
"fan_in_fan_out": false,
|
| 13 |
+
"inference_mode": true,
|
| 14 |
+
"init_lora_weights": true,
|
| 15 |
+
"kasa_config": null,
|
| 16 |
+
"layer_replication": null,
|
| 17 |
+
"layers_pattern": null,
|
| 18 |
+
"layers_to_transform": null,
|
| 19 |
+
"loftq_config": {},
|
| 20 |
+
"lora_alpha": 8,
|
| 21 |
+
"lora_bias": false,
|
| 22 |
+
"lora_dropout": 0.0,
|
| 23 |
+
"lora_ga_config": null,
|
| 24 |
+
"megatron_config": null,
|
| 25 |
+
"megatron_core": "megatron.core",
|
| 26 |
+
"modules_to_save": null,
|
| 27 |
+
"monteclora_config": null,
|
| 28 |
+
"peft_type": "LORA",
|
| 29 |
+
"peft_version": "0.21.2",
|
| 30 |
+
"qalora_group_size": 16,
|
| 31 |
+
"r": 4,
|
| 32 |
+
"rank_pattern": {},
|
| 33 |
+
"revision": "73500c80fcb6c8bea3e830c21d5d2920249a0e1a",
|
| 34 |
+
"target_modules": [
|
| 35 |
+
"q_proj",
|
| 36 |
+
"v_proj"
|
| 37 |
+
],
|
| 38 |
+
"target_parameters": null,
|
| 39 |
+
"task_type": "CAUSAL_LM",
|
| 40 |
+
"trainable_token_indices": null,
|
| 41 |
+
"use_bdlora": null,
|
| 42 |
+
"use_dora": false,
|
| 43 |
+
"use_qalora": false,
|
| 44 |
+
"use_rslora": false,
|
| 45 |
+
"velora_config": null
|
| 46 |
+
}
|
adapter_model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cad9f72b601182473a88633ad5e68d703eb5f93a21360639276be7871a4b0fba
|
| 3 |
+
size 3852744
|
provenance-manifest.json
ADDED
|
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"base_model": "CohereLabs/tiny-aya-fire",
|
| 3 |
+
"base_revision": "73500c80fcb6c8bea3e830c21d5d2920249a0e1a",
|
| 4 |
+
"adapter_count": 1,
|
| 5 |
+
"artifact_weight_hashes_unchanged": true,
|
| 6 |
+
"original_configs_preserved": true,
|
| 7 |
+
"config_changes_limited_to_portability_and_revision": true,
|
| 8 |
+
"base_weights_or_tokenizer_assets_included": false,
|
| 9 |
+
"new_model_execution_or_provider_mutation": false,
|
| 10 |
+
"adapters": [
|
| 11 |
+
{
|
| 12 |
+
"model_id": "state-only-sft",
|
| 13 |
+
"model_role": "trained_state_classification_adapter",
|
| 14 |
+
"role_description": "State-only classifier; 864 actual updates and frozen English classification evaluation.",
|
| 15 |
+
"base_model": "CohereLabs/tiny-aya-fire",
|
| 16 |
+
"base_model_revision": "73500c80fcb6c8bea3e830c21d5d2920249a0e1a",
|
| 17 |
+
"adapter_weight_sha256": "cad9f72b601182473a88633ad5e68d703eb5f93a21360639276be7871a4b0fba",
|
| 18 |
+
"adapter_bytes": 3852744,
|
| 19 |
+
"adapter_weight_bytes_unchanged": true,
|
| 20 |
+
"new_training_or_model_execution": false,
|
| 21 |
+
"source_archive_chain": [
|
| 22 |
+
"outputs/HF-state-training-retry-results.zip"
|
| 23 |
+
],
|
| 24 |
+
"outer_source_archive_sha256": "a9cc7b6af8264d977b3a769aa86334ad41b22f14aa8829880f1093c301a41688",
|
| 25 |
+
"source_adapter_member": "state-training/20261106T131307463330Z/adapter/adapter_model.safetensors",
|
| 26 |
+
"source_config_member": "state-training/20261106T131307463330Z/adapter/adapter_config.json",
|
| 27 |
+
"original_config_sha256": "44c9f947910559bfb3c56557badf4b0d393908c0c9d89a78beadf35e575b8333",
|
| 28 |
+
"normalized_config_sha256": "89eec08b76d40c589d69e290ef5069e724e4b6921bb41c9551bcff70bcf647ce",
|
| 29 |
+
"config_changes": {
|
| 30 |
+
"base_model_name_or_path": {
|
| 31 |
+
"before": "/content/evidencelock-gpu/hf-home/hub/models--CohereLabs--tiny-aya-fire/snapshots/73500c80fcb6c8bea3e830c21d5d2920249a0e1a",
|
| 32 |
+
"after": "CohereLabs/tiny-aya-fire"
|
| 33 |
+
},
|
| 34 |
+
"revision": {
|
| 35 |
+
"before": null,
|
| 36 |
+
"after": "73500c80fcb6c8bea3e830c21d5d2920249a0e1a"
|
| 37 |
+
}
|
| 38 |
+
},
|
| 39 |
+
"config_change_purpose": "Portable base-model identifier and explicit revision pin; no other configuration or weight changes.",
|
| 40 |
+
"source_provenance_members": [
|
| 41 |
+
"compact-state-diagnostic/latest-result.json",
|
| 42 |
+
"fresh-model-access-result.json",
|
| 43 |
+
"fresh-recovery-manifest.json",
|
| 44 |
+
"gpu-environment-result.json",
|
| 45 |
+
"state-training/20261106T131307463330Z/adapter/adapter_config.json",
|
| 46 |
+
"state-training/20261106T131307463330Z/result.json",
|
| 47 |
+
"state-training/latest-result.json"
|
| 48 |
+
]
|
| 49 |
+
}
|
| 50 |
+
]
|
| 51 |
+
}
|
provenance/config-normalization.json
ADDED
|
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_id": "state-only-sft",
|
| 3 |
+
"model_role": "trained_state_classification_adapter",
|
| 4 |
+
"role_description": "State-only classifier; 864 actual updates and frozen English classification evaluation.",
|
| 5 |
+
"base_model": "CohereLabs/tiny-aya-fire",
|
| 6 |
+
"base_model_revision": "73500c80fcb6c8bea3e830c21d5d2920249a0e1a",
|
| 7 |
+
"adapter_weight_sha256": "cad9f72b601182473a88633ad5e68d703eb5f93a21360639276be7871a4b0fba",
|
| 8 |
+
"adapter_bytes": 3852744,
|
| 9 |
+
"adapter_weight_bytes_unchanged": true,
|
| 10 |
+
"new_training_or_model_execution": false,
|
| 11 |
+
"source_archive_chain": [
|
| 12 |
+
"outputs/HF-state-training-retry-results.zip"
|
| 13 |
+
],
|
| 14 |
+
"outer_source_archive_sha256": "a9cc7b6af8264d977b3a769aa86334ad41b22f14aa8829880f1093c301a41688",
|
| 15 |
+
"source_adapter_member": "state-training/20261106T131307463330Z/adapter/adapter_model.safetensors",
|
| 16 |
+
"source_config_member": "state-training/20261106T131307463330Z/adapter/adapter_config.json",
|
| 17 |
+
"original_config_sha256": "44c9f947910559bfb3c56557badf4b0d393908c0c9d89a78beadf35e575b8333",
|
| 18 |
+
"normalized_config_sha256": "89eec08b76d40c589d69e290ef5069e724e4b6921bb41c9551bcff70bcf647ce",
|
| 19 |
+
"config_changes": {
|
| 20 |
+
"base_model_name_or_path": {
|
| 21 |
+
"before": "/content/evidencelock-gpu/hf-home/hub/models--CohereLabs--tiny-aya-fire/snapshots/73500c80fcb6c8bea3e830c21d5d2920249a0e1a",
|
| 22 |
+
"after": "CohereLabs/tiny-aya-fire"
|
| 23 |
+
},
|
| 24 |
+
"revision": {
|
| 25 |
+
"before": null,
|
| 26 |
+
"after": "73500c80fcb6c8bea3e830c21d5d2920249a0e1a"
|
| 27 |
+
}
|
| 28 |
+
},
|
| 29 |
+
"config_change_purpose": "Portable base-model identifier and explicit revision pin; no other configuration or weight changes.",
|
| 30 |
+
"source_provenance_members": [
|
| 31 |
+
"compact-state-diagnostic/latest-result.json",
|
| 32 |
+
"fresh-model-access-result.json",
|
| 33 |
+
"fresh-recovery-manifest.json",
|
| 34 |
+
"gpu-environment-result.json",
|
| 35 |
+
"state-training/20261106T131307463330Z/adapter/adapter_config.json",
|
| 36 |
+
"state-training/20261106T131307463330Z/result.json",
|
| 37 |
+
"state-training/latest-result.json"
|
| 38 |
+
]
|
| 39 |
+
}
|
provenance/original_adapter_config.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"alora_invocation_tokens": null,
|
| 3 |
+
"alpha_pattern": {},
|
| 4 |
+
"arrow_config": null,
|
| 5 |
+
"auto_mapping": null,
|
| 6 |
+
"base_model_name_or_path": "/content/evidencelock-gpu/hf-home/hub/models--CohereLabs--tiny-aya-fire/snapshots/73500c80fcb6c8bea3e830c21d5d2920249a0e1a",
|
| 7 |
+
"bias": "none",
|
| 8 |
+
"corda_config": null,
|
| 9 |
+
"ensure_weight_tying": false,
|
| 10 |
+
"eva_config": null,
|
| 11 |
+
"exclude_modules": null,
|
| 12 |
+
"fan_in_fan_out": false,
|
| 13 |
+
"inference_mode": true,
|
| 14 |
+
"init_lora_weights": true,
|
| 15 |
+
"kasa_config": null,
|
| 16 |
+
"layer_replication": null,
|
| 17 |
+
"layers_pattern": null,
|
| 18 |
+
"layers_to_transform": null,
|
| 19 |
+
"loftq_config": {},
|
| 20 |
+
"lora_alpha": 8,
|
| 21 |
+
"lora_bias": false,
|
| 22 |
+
"lora_dropout": 0.0,
|
| 23 |
+
"lora_ga_config": null,
|
| 24 |
+
"megatron_config": null,
|
| 25 |
+
"megatron_core": "megatron.core",
|
| 26 |
+
"modules_to_save": null,
|
| 27 |
+
"monteclora_config": null,
|
| 28 |
+
"peft_type": "LORA",
|
| 29 |
+
"peft_version": "0.21.2",
|
| 30 |
+
"qalora_group_size": 16,
|
| 31 |
+
"r": 4,
|
| 32 |
+
"rank_pattern": {},
|
| 33 |
+
"revision": null,
|
| 34 |
+
"target_modules": [
|
| 35 |
+
"q_proj",
|
| 36 |
+
"v_proj"
|
| 37 |
+
],
|
| 38 |
+
"target_parameters": null,
|
| 39 |
+
"task_type": "CAUSAL_LM",
|
| 40 |
+
"trainable_token_indices": null,
|
| 41 |
+
"use_bdlora": null,
|
| 42 |
+
"use_dora": false,
|
| 43 |
+
"use_qalora": false,
|
| 44 |
+
"use_rslora": false,
|
| 45 |
+
"velora_config": null
|
| 46 |
+
}
|