diff --git a/e28/S1-E28-C0-syn/DONE b/e28/S1-E28-C0-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..68ba314868b076e46dd9dbaf3bbebfc2ae09941b --- /dev/null +++ b/e28/S1-E28-C0-syn/DONE @@ -0,0 +1 @@ +2026-10-20 06:33:13 \ No newline at end of file diff --git a/e28/S1-E28-C0-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-C0-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-C0-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-C0-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-C0-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..6bd44e5abd21fe14f480ba28f6e1198fcdc2f290 --- /dev/null +++ b/e28/S1-E28-C0-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "up_proj", + "o_proj", + "v_proj", + "in_proj_qkv", + "k_proj", + "gate_proj", + "out_proj", + "q_proj", + "in_proj_z", + "down_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-C0-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-C0-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..148ce1df81a12d0d951c5a4e1a80c26eb7021489 --- /dev/null +++ b/e28/S1-E28-C0-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea15bc4fc52ad521385fee6a115227c0b50e9a7164e2578a47cc48e63362d083 +size 160486456 diff --git a/e28/S1-E28-C0-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-C0-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-C0-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-C0-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-C0-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dc269c66c9e5a9c8518e1d1fa9c25f6eedcfb2c1 --- /dev/null +++ b/e28/S1-E28-C0-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4805bbcf619946a680b929f9c21e8dbd0d9ba163167dffb7cc13dd8182303fce +size 176641832 diff --git a/e28/S1-E28-C0-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-C0-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..1a6600c30751c6fc54419522034fee4e649d4354 --- /dev/null +++ b/e28/S1-E28-C0-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.0004019341722596437, + -0.00040218859794549644, + -0.0023606743197888136, + -0.0017429249128326774, + -0.0005036078509874642, + 0.0018798642558977008, + 0.0014552718494087458, + -0.002158828778192401, + 8.943251305026934e-05, + -7.114288746379316e-05, + -5.446747763926396e-07, + 0.0007199719548225403, + -0.0017304084030911326, + -0.003423966933041811, + -0.004143770784139633, + -0.005592531058937311, + -0.005568793509155512, + -0.00443001464009285, + -0.0049525978974998, + -0.005071173422038555, + -0.003785105422139168, + -0.003439312567934394, + -0.005798427853733301, + -0.0035148358438163996 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-C0-syn/ckpt-16000/calibration.json b/e28/S1-E28-C0-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..96d27f91b14e678e8b50600753fc81e3d0d22c55 --- /dev/null +++ b/e28/S1-E28-C0-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.8210398824005378, + "choice": 1.0076231342175517, + "score": 0.9955852301156168 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.8210398824005378, + "ce_before": 0.5733858942985535, + "ce_after": 0.5705401301383972 + }, + "choice": { + "n": 60, + "T": 1.0076231342175517, + "ce_before": 0.6707404851913452, + "ce_after": 0.6707300543785095 + }, + "score": { + "n": 53, + "T": 0.9955852301156168, + "ce_before": 1.055969476699829, + "ce_after": 1.0559628009796143 + } + }, + "source": "models/e28/S1-E28-C0-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-C0-syn/ckpt-16000/head.safetensors b/e28/S1-E28-C0-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f8eed5bd64054758567dfefe8842e84755863bf0 --- /dev/null +++ b/e28/S1-E28-C0-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6c7b7a054dfd9924cb58a4007d7caea6920434919ef35e46b9e5daf7c0fefb66 +size 393464 diff --git a/e28/S1-E28-C0-syn/ckpt-16000/judge_config.json b/e28/S1-E28-C0-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..bcab2955b6d402bc3b6930a93e213eef50fb0ff6 --- /dev/null +++ b/e28/S1-E28-C0-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-C0-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e19/E19-b-n16k", + "data_manifest": { + "arm": "E19-b-n16k", + "name": null, + "train_sha256": "9e31412d9aac0064859cdf9aae1636242d1a9dbfc3ab398a1f5ec10da1e7a50e", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9808867, + "tok_per_s": 3237.0095425583345, + "rows_per_s": 5.558003251258509, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5733858942985535, + "noul_T": 0.8210398824005378, + "choice_ce": 0.6707404851913452, + "choice_T": 1.0076231342175517, + "score_ce": 1.055969476699829, + "score_T": 0.9955852301156168, + "noul_auc": 0.7198933333333333, + "noul_acc": 0.699, + "noul_f1": 0.7794871794871795, + "noul_nll": 0.5898938082531094, + "noul_ece": 0.052008005946874616, + "noul_brier": 0.20208435675807834, + "noul_sharp": 0.1651003509759903, + "entropy": 0.6122925281524658 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-C0-syn/train_log.jsonl b/e28/S1-E28-C0-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..160148398d18f8b4e1136c5632a63f1a951fd607 --- /dev/null +++ b/e28/S1-E28-C0-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "05:42:42", "event": "start", "data": "data/e19/E19-b-n16k", "data_manifest": {"arm": "E19-b-n16k", "name": null, "train_sha256": "9e31412d9aac0064859cdf9aae1636242d1a9dbfc3ab398a1f5ec10da1e7a50e", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.794634580612183} +{"time": "05:44:34", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 1742.0688985934723, "loss": 0.6418524533510208, "entropy": 0.6674938574433327, "sharp": 0.09449283499270678} +{"time": "05:45:28", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 2276.391236296654, "loss": 0.6543252449482679, "entropy": 0.6451239045709372, "sharp": 0.11809633672237396} +{"time": "05:46:27", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 2592.524926228113, "loss": 0.7097902707755566, "entropy": 0.6600528657436371, "sharp": 0.11035785730928183} +{"time": "05:47:22", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 2768.586496905704, "loss": 0.6805907804518938, "entropy": 0.6736207865178585, "sharp": 0.08694715239107609} +{"time": "05:48:18", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 2889.323608493551, "loss": 0.6551038809120655, "entropy": 0.6535464469343424, "sharp": 0.10296877473592758} +{"time": "05:49:14", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 2977.4619186367863, "loss": 0.6448147781193256, "entropy": 0.6451537758111954, "sharp": 0.12875191681087017} +{"time": "05:50:12", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3036.6014164640724, "loss": 0.647441454231739, "entropy": 0.683087445795536, "sharp": 0.057668830268085} +{"time": "05:51:08", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3084.627950729468, "loss": 0.7112917471677065, "entropy": 0.637435607612133, "sharp": 0.1373877078294754} +{"time": "05:52:04", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3124.1413080678926, "loss": 0.6655900441110134, "entropy": 0.6652580164372921, "sharp": 0.10431478638201952} +{"time": "05:53:00", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3152.110435086271, "loss": 0.6226409263908863, "entropy": 0.6681161969900131, "sharp": 0.09350143652409315} +{"time": "05:53:25", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.6027932167053223, "noul_T": 0.8047402751192091, "noul_auc": 0.7031611280822265, "noul_acc": 0.688, "noul_f1": 0.75625, "noul_nll": 0.6076746896356344, "noul_ece": 0.06157691490650178, "noul_brier": 0.2100360241730981, "noul_sharp": 0.14764730632305145, "entropy": 0.6277173757553101} +{"time": "05:54:23", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3068.6206931182887, "loss": 0.684677317738533, "entropy": 0.6516259256750345, "sharp": 0.11634835135191679} +{"time": "05:55:19", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3089.830183059434, "loss": 0.6259176284074783, "entropy": 0.6395060457289219, "sharp": 0.14079864881932735} +{"time": "05:56:16", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3116.7881936735016, "loss": 0.6442239172756672, "entropy": 0.6784866154193878, "sharp": 0.06614390201866627} +{"time": "05:57:14", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3139.759900493683, "loss": 0.6844842508435249, "entropy": 0.64948820322752, "sharp": 0.13435495272278786} +{"time": "05:58:09", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3160.5145456206246, "loss": 0.6813701651990414, "entropy": 0.668898917734623, "sharp": 0.08269481174647808} +{"time": "05:59:06", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3179.978036598646, "loss": 0.7068287283182144, "entropy": 0.6260516047477722, "sharp": 0.15484808944165707} +{"time": "06:00:04", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3196.797926526836, "loss": 0.6369537115097046, "entropy": 0.6493389010429382, "sharp": 0.11900342628359795} +{"time": "06:01:00", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3211.968891141825, "loss": 0.6736315377056599, "entropy": 0.6538582667708397, "sharp": 0.11945641413331032} +{"time": "06:01:56", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3223.8353887126264, "loss": 0.6503072865307331, "entropy": 0.6558262407779694, "sharp": 0.10804163292050362} +{"time": "06:02:53", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3235.046518412555, "loss": 0.6812480166554451, "entropy": 0.6733076274394989, "sharp": 0.07814495917409658} +{"time": "06:03:13", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.625215470790863, "noul_T": 0.6322796617575723, "noul_auc": 0.7433798585677879, "noul_acc": 0.66, "noul_f1": 0.6875, "noul_nll": 0.6210226207077503, "noul_ece": 0.14127235460281373, "noul_brier": 0.21537573031776094, "noul_sharp": 0.11072076112031937, "entropy": 0.6588683724403381} +{"time": "06:04:09", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3195.1610605836995, "loss": 0.6731307283043861, "entropy": 0.6324892044067383, "sharp": 0.14314057677984238} +{"time": "06:05:04", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3207.7340288322866, "loss": 0.7052969560027122, "entropy": 0.6550907492637634, "sharp": 0.11679293122142553} +{"time": "06:06:01", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3218.423092980036, "loss": 0.6667746938765049, "entropy": 0.6606782749295235, "sharp": 0.10835204273462296} +{"time": "06:06:58", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3228.7448099209337, "loss": 0.6714807525277138, "entropy": 0.6531186066567898, "sharp": 0.11655727680772543} +{"time": "06:07:54", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3238.0592523581136, "loss": 0.6987750642001629, "entropy": 0.6320639178156853, "sharp": 0.15557408891618252} +{"time": "06:08:50", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3245.5095129602087, "loss": 0.653316430747509, "entropy": 0.6700648739933968, "sharp": 0.0919972499832511} +{"time": "06:09:49", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3255.199106196958, "loss": 0.640799886547029, "entropy": 0.6467154920101166, "sharp": 0.13283592090010643} +{"time": "06:10:46", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3262.485544814, "loss": 0.6944353505969048, "entropy": 0.6748707816004753, "sharp": 0.08099196664988995} +{"time": "06:11:43", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3270.3215016883805, "loss": 0.6646610796451569, "entropy": 0.6652002595365047, "sharp": 0.0954577811062336} +{"time": "06:12:41", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3276.736460884371, "loss": 0.6731904521584511, "entropy": 0.6602216698229313, "sharp": 0.11017826851457357} +{"time": "06:13:01", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.597034752368927, "noul_T": 0.6421206855787178, "noul_auc": 0.7323589475283117, "noul_acc": 0.718, "noul_f1": 0.7860394537177542, "noul_nll": 0.5973743291497231, "noul_ece": 0.08663646721839906, "noul_brier": 0.2042407758885801, "noul_sharp": 0.1332443803548813, "entropy": 0.6418294906616211} +{"time": "06:13:56", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3247.7827177374384, "loss": 0.6042209938168526, "entropy": 0.6657007820904255, "sharp": 0.10047882981598377} +{"time": "06:14:53", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3254.074569995877, "loss": 0.672166720032692, "entropy": 0.6612759456038475, "sharp": 0.09925397951155901} +{"time": "06:15:51", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3260.7679436398503, "loss": 0.6972816735506058, "entropy": 0.643512986600399, "sharp": 0.12919290456920862} +{"time": "06:16:45", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3266.4804837555394, "loss": 0.6724905520677567, "entropy": 0.6493063047528267, "sharp": 0.11664058640599251} +{"time": "06:17:42", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3272.1660910371143, "loss": 0.6453768238425255, "entropy": 0.6563559770584106, "sharp": 0.11054976005107164} +{"time": "06:18:38", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3277.0396115841886, "loss": 0.6560497954487801, "entropy": 0.635651208460331, "sharp": 0.14939494617283344} +{"time": "06:19:35", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3282.9871352353107, "loss": 0.686661321669817, "entropy": 0.6401583924889565, "sharp": 0.13821567595005035} +{"time": "06:20:32", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3287.3508514962205, "loss": 0.6733216270804405, "entropy": 0.6461369395256042, "sharp": 0.12509455811232328} +{"time": "06:21:29", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3293.099189016152, "loss": 0.6431321948766708, "entropy": 0.6418887823820114, "sharp": 0.1216413863003254} +{"time": "06:22:23", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3296.8323002762377, "loss": 0.6882326304912567, "entropy": 0.6303502470254898, "sharp": 0.14134303573518991} +{"time": "06:22:43", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5883752703666687, "noul_T": 0.8063948425248696, "noul_auc": 0.7352956068252012, "noul_acc": 0.722, "noul_f1": 0.7851622874806801, "noul_nll": 0.5876846368610859, "noul_ece": 0.05814056313037875, "noul_brier": 0.20021823288795024, "noul_sharp": 0.1605500727891922, "entropy": 0.6165220141410828} +{"time": "06:23:40", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3273.217062149899, "loss": 0.6581775918602943, "entropy": 0.6659778021275997, "sharp": 0.09172998555004597} +{"time": "06:24:35", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3278.0760739602315, "loss": 0.6624073907732964, "entropy": 0.6563186943531036, "sharp": 0.10841984767466784} +{"time": "06:25:33", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3282.822632861716, "loss": 0.668071087449789, "entropy": 0.646204873919487, "sharp": 0.12354523129761219} +{"time": "06:26:29", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3287.3099151168462, "loss": 0.6683605443686247, "entropy": 0.6525228191167116, "sharp": 0.11701337993144989} +{"time": "06:27:26", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3291.6826129417955, "loss": 0.6479352544993162, "entropy": 0.6605901643633842, "sharp": 0.09802819322794676} +{"time": "06:28:22", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3295.468233519694, "loss": 0.6506122052669525, "entropy": 0.6533954069018364, "sharp": 0.11456773895770311} +{"time": "06:29:20", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3299.817314706129, "loss": 0.7218317352235317, "entropy": 0.6592727117240429, "sharp": 0.10681359004229307} +{"time": "06:30:16", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3303.496486455061, "loss": 0.6903524547815323, "entropy": 0.6571662090718746, "sharp": 0.10609687305986881} +{"time": "06:31:10", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3306.3984164704566, "loss": 0.5719492360949516, "entropy": 0.6424555853009224, "sharp": 0.13239238318055868} +{"time": "06:32:06", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3309.895620939873, "loss": 0.6167443245649338, "entropy": 0.6454180404543877, "sharp": 0.12842741049826145} +{"time": "06:32:26", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5884483456611633, "noul_T": 0.798413540070553, "noul_auc": 0.7332534471992033, "noul_acc": 0.72, "noul_f1": 0.7859327217125383, "noul_nll": 0.5891514290720224, "noul_ece": 0.061191798210144054, "noul_brier": 0.200934913549001, "noul_sharp": 0.15952661633491516, "entropy": 0.6176946759223938} +{"time": "06:33:13", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9808867, "tok_per_s": 3237.0095425583345, "rows_per_s": 5.558003251258509, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5733858942985535, "noul_T": 0.8210398824005378, "choice_ce": 0.6707404851913452, "choice_T": 1.0076231342175517, "score_ce": 1.055969476699829, "score_T": 0.9955852301156168, "noul_auc": 0.7198933333333333, "noul_acc": 0.699, "noul_f1": 0.7794871794871795, "noul_nll": 0.5898938082531094, "noul_ece": 0.052008005946874616, "noul_brier": 0.20208435675807834, "noul_sharp": 0.1651003509759903, "entropy": 0.6122925281524658}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-E18A050-rt-syn/DONE b/e28/S1-E28-E18A050-rt-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..e4cada86b2785e5ff515eeda9461055c54f3aadb --- /dev/null +++ b/e28/S1-E28-E18A050-rt-syn/DONE @@ -0,0 +1 @@ +2026-10-21 06:54:56 \ No newline at end of file diff --git a/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d97f67fba27218f63eba406b52434759f889d42 --- /dev/null +++ b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "o_proj", + "k_proj", + "in_proj_z", + "out_proj", + "q_proj", + "v_proj", + "up_proj", + "down_proj", + "in_proj_qkv", + "gate_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7c60e051a1f261fab7d1829cac9e67e9bdcb53bc --- /dev/null +++ b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d32e8cade04cb62e764ce8960d6d4340906a030b264367825ba9b79963a9929 +size 160486456 diff --git a/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..cf5730b97f920e8246de52b0d15a31680e185428 --- /dev/null +++ b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb2307e1988ab4b5c8524539ee27ea8a645102f2a10ab1c486095071391da9a6 +size 176641832 diff --git a/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..43e8c398e745e3ce130151d24c8c98f8c5e6a7ce --- /dev/null +++ b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.00039831086178310215, + -0.00039857125375419855, + -0.0030545094050467014, + -0.0019210963509976864, + -0.0006272284663282335, + 0.0020584675949066877, + 0.0012770768953487277, + -0.003080836497247219, + -0.00024181828484870493, + -0.0002333867159904912, + 0.00029039953369647264, + 0.0007457923493348062, + 0.00015367088781204075, + -0.0038240239955484867, + -0.00451631611213088, + -0.005286028143018484, + -0.006027279421687126, + -0.004458629060536623, + -0.004973826929926872, + -0.005131916608661413, + -0.0040062107145786285, + -0.003536548698320985, + -0.005771616939455271, + -0.003572249785065651 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-E18A050-rt-syn/ckpt-16000/calibration.json b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..c9c78392e652bec4ba4434ef3581769c97852456 --- /dev/null +++ b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.8495371650607102, + "choice": 1.0560892190772713, + "score": 0.960112119036335 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.8495371650607102, + "ce_before": 0.5595800280570984, + "ce_after": 0.5574237108230591 + }, + "choice": { + "n": 60, + "T": 1.0560892190772713, + "ce_before": 0.6842275261878967, + "ce_after": 0.683573842048645 + }, + "score": { + "n": 53, + "T": 0.960112119036335, + "ce_before": 1.068756341934204, + "ce_after": 1.0682735443115234 + } + }, + "source": "models/e28/S1-E28-E18A050-rt-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-E18A050-rt-syn/ckpt-16000/head.safetensors b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7ca8e05d2985b45a764bfe81b39944818e67fb11 --- /dev/null +++ b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04f1e3aebba797e65e6f251cf88d134d9cb90ba13166ecaa4ef8d3113bfbdf1d +size 393464 diff --git a/e28/S1-E28-E18A050-rt-syn/ckpt-16000/judge_config.json b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2958674ac3fbda7ab4efd56431076a399b2f728a --- /dev/null +++ b/e28/S1-E28-E18A050-rt-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-E18A050-rt-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-E18A050-rt", + "data_manifest": { + "arm": "G-d512-n16k", + "name": null, + "train_sha256": "ef94315960694ce58bdc6d61b8665b7bf46f4e2996b9e82503e1d17bb87a14b7", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9113372, + "tok_per_s": 3214.371023602123, + "rows_per_s": 5.940329960631729, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5595800280570984, + "noul_T": 0.8495371650607102, + "choice_ce": 0.6842275261878967, + "choice_T": 1.0560892190772713, + "score_ce": 1.068756341934204, + "score_T": 0.960112119036335, + "noul_auc": 0.7447338666666667, + "noul_acc": 0.695, + "noul_f1": 0.7677075399847677, + "noul_nll": 0.5717920997068286, + "noul_ece": 0.04427964748442174, + "noul_brier": 0.19495265463736913, + "noul_sharp": 0.1802748143672943, + "entropy": 0.5956453680992126 + }, + "dropped_long": 0, + "peak_gib": 18.284047603607178 + } +} \ No newline at end of file diff --git a/e28/S1-E28-E18A050-rt-syn/train_log.jsonl b/e28/S1-E28-E18A050-rt-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..aaab771612aee639c8377949bc326e510917f6a2 --- /dev/null +++ b/e28/S1-E28-E18A050-rt-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "06:07:41", "event": "start", "data": "data/e28/E28-E18A050-rt", "data_manifest": {"arm": "G-d512-n16k", "name": null, "train_sha256": "ef94315960694ce58bdc6d61b8665b7bf46f4e2996b9e82503e1d17bb87a14b7", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.799657821655273} +{"time": "06:08:35", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3335.8686490784153, "loss": 0.6002982165664434, "entropy": 0.5942575950175524, "sharp": 0.19537658989429474} +{"time": "06:09:29", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3347.89659647421, "loss": 0.6112224832177162, "entropy": 0.59024703130126, "sharp": 0.2019644845277071} +{"time": "06:10:23", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3353.529258976685, "loss": 0.6076782438904047, "entropy": 0.529285728931427, "sharp": 0.24768979381769896} +{"time": "06:11:16", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3359.1094743060803, "loss": 0.5808004774153233, "entropy": 0.6097748354077339, "sharp": 0.17353096511214972} +{"time": "06:12:11", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3369.3068844360805, "loss": 0.5585350655019283, "entropy": 0.5976961776614189, "sharp": 0.1846163496375084} +{"time": "06:13:04", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3375.8048401322844, "loss": 0.5941595509648323, "entropy": 0.6318124160170555, "sharp": 0.154153554700315} +{"time": "06:13:58", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3375.8535966310665, "loss": 0.5850520320236683, "entropy": 0.5771617069840431, "sharp": 0.20421438571065664} +{"time": "06:14:52", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3377.0916069751543, "loss": 0.6283475775271654, "entropy": 0.601615758612752, "sharp": 0.18012536317110062} +{"time": "06:15:47", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3380.7041022056424, "loss": 0.6425563246011734, "entropy": 0.594768863171339, "sharp": 0.17997646797448397} +{"time": "06:16:41", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3380.585620079485, "loss": 0.5740486830472946, "entropy": 0.6155685745179653, "sharp": 0.16669677570462227} +{"time": "06:17:01", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.5806242823600769, "noul_T": 0.9017193599054513, "noul_auc": 0.7404600766231794, "noul_acc": 0.708, "noul_f1": 0.7652733118971061, "noul_nll": 0.5830994355976582, "noul_ece": 0.06945477390289306, "noul_brier": 0.1993663816604694, "noul_sharp": 0.1820717602968216, "entropy": 0.5966274738311768} +{"time": "06:17:54", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3274.935834648614, "loss": 0.6264102570712566, "entropy": 0.6102959588170052, "sharp": 0.16732802893966436} +{"time": "06:18:50", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3287.0189864821386, "loss": 0.5803467445075512, "entropy": 0.5993821527808905, "sharp": 0.17582901380956173} +{"time": "06:19:47", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3297.0851454234808, "loss": 0.5702192187309265, "entropy": 0.5612410046160221, "sharp": 0.21354103833436966} +{"time": "06:20:41", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3302.716488321634, "loss": 0.6200642455369234, "entropy": 0.6473408713936806, "sharp": 0.12580503337085247} +{"time": "06:21:36", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3308.2881403457145, "loss": 0.6428153924643993, "entropy": 0.6326981969177723, "sharp": 0.14672968070954084} +{"time": "06:22:28", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3312.8328430870833, "loss": 0.6297238897532225, "entropy": 0.5767229981720448, "sharp": 0.19916999712586403} +{"time": "06:23:23", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3318.1738265316453, "loss": 0.6130420938134193, "entropy": 0.6221097409725189, "sharp": 0.15260346792638302} +{"time": "06:24:17", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3322.7193927372523, "loss": 0.5934165082871914, "entropy": 0.6152561269700527, "sharp": 0.16756878979504108} +{"time": "06:25:11", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3324.589217224558, "loss": 0.6296769827604294, "entropy": 0.6074256487190723, "sharp": 0.16790608409792185} +{"time": "06:26:05", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3328.6099107837185, "loss": 0.5907804258167744, "entropy": 0.6081702299416065, "sharp": 0.18031224142760038} +{"time": "06:26:25", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5784547328948975, "noul_T": 0.8328414215411776, "noul_auc": 0.7448313108639516, "noul_acc": 0.704, "noul_f1": 0.7620578778135049, "noul_nll": 0.5811050497889518, "noul_ece": 0.06479078799486161, "noul_brier": 0.19810341077413438, "noul_sharp": 0.17559659481048584, "entropy": 0.6053183078765869} +{"time": "06:27:20", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3274.7058554089595, "loss": 0.586168397217989, "entropy": 0.5733136478811502, "sharp": 0.20715579017996788} +{"time": "06:28:13", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3278.7195425829236, "loss": 0.6014519557356834, "entropy": 0.593541793525219, "sharp": 0.18261277489364147} +{"time": "06:29:06", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3282.2559446384985, "loss": 0.5910805948078632, "entropy": 0.5957015547901392, "sharp": 0.19531946256756783} +{"time": "06:29:59", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3285.9311897875646, "loss": 0.5925949309021235, "entropy": 0.5877530910074711, "sharp": 0.19521856494247913} +{"time": "06:30:57", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3288.701294556637, "loss": 0.6153495758771896, "entropy": 0.6061336323618889, "sharp": 0.1711768750101328} +{"time": "06:31:48", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3291.160656994244, "loss": 0.5898217596113682, "entropy": 0.6045204438269138, "sharp": 0.16672593727707863} +{"time": "06:32:41", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3294.217945249048, "loss": 0.6414357991889119, "entropy": 0.6187189966440201, "sharp": 0.15950102172791958} +{"time": "06:33:34", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3298.173177428458, "loss": 0.6342934807762504, "entropy": 0.6112399101257324, "sharp": 0.17046577669680119} +{"time": "06:34:29", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3301.4133787114624, "loss": 0.617023091763258, "entropy": 0.5880597829818726, "sharp": 0.18076064437627792} +{"time": "06:35:24", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3304.863377969232, "loss": 0.5869416482746601, "entropy": 0.5788372401148081, "sharp": 0.1744002364575863} +{"time": "06:35:44", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5695884227752686, "noul_T": 0.8970043922903317, "noul_auc": 0.7486624698317328, "noul_acc": 0.706, "noul_f1": 0.7692307692307693, "noul_nll": 0.5728995004072786, "noul_ece": 0.03249754613637924, "noul_brier": 0.19544740658234394, "noul_sharp": 0.18595445156097412, "entropy": 0.586853563785553} +{"time": "06:36:36", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3268.4624207100774, "loss": 0.5773440040647984, "entropy": 0.6028434857726097, "sharp": 0.16967926733195782} +{"time": "06:37:27", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3271.3165014306455, "loss": 0.6374371545389295, "entropy": 0.6155262598767877, "sharp": 0.16538868471980095} +{"time": "06:38:21", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3275.03692884936, "loss": 0.5824361257255077, "entropy": 0.5844896733760834, "sharp": 0.18815483525395393} +{"time": "06:39:15", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3278.093736059273, "loss": 0.6290938165038824, "entropy": 0.6002766862511635, "sharp": 0.17840714938938618} +{"time": "06:40:06", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3281.1930558265444, "loss": 0.6121017262339592, "entropy": 0.61546690762043, "sharp": 0.14994669798761606} +{"time": "06:40:59", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3284.7178698010935, "loss": 0.6235643997788429, "entropy": 0.598175898194313, "sharp": 0.18592064268887043} +{"time": "06:41:52", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3288.22198246336, "loss": 0.6016076346859336, "entropy": 0.5771215986460447, "sharp": 0.20257375203073025} +{"time": "06:42:47", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3290.5559815850797, "loss": 0.6171148121356964, "entropy": 0.6074099205434322, "sharp": 0.1688613723963499} +{"time": "06:43:43", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3294.047448880535, "loss": 0.6298759374767542, "entropy": 0.6090978160500526, "sharp": 0.16564554441720247} +{"time": "06:44:35", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3295.362348553038, "loss": 0.6032796315848827, "entropy": 0.607449185103178, "sharp": 0.16136501543223858} +{"time": "06:44:55", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5710271000862122, "noul_T": 0.8856563235718324, "noul_auc": 0.7491350356955999, "noul_acc": 0.712, "noul_f1": 0.7770897832817337, "noul_nll": 0.5730964285135269, "noul_ece": 0.03887997013330461, "noul_brier": 0.19541467243427132, "noul_sharp": 0.18427079916000366, "entropy": 0.590028703212738} +{"time": "06:45:48", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3268.2654171836684, "loss": 0.6083678230643272, "entropy": 0.602668821811676, "sharp": 0.17243992909789085} +{"time": "06:46:40", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3269.8583601611954, "loss": 0.6399169377982616, "entropy": 0.6169490301981568, "sharp": 0.15903472062200308} +{"time": "06:47:35", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3272.142937401324, "loss": 0.612063997425139, "entropy": 0.5993046313524246, "sharp": 0.17189513705670834} +{"time": "06:48:27", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3274.5345470432685, "loss": 0.6128622889518738, "entropy": 0.6048808097839355, "sharp": 0.16242566797882318} +{"time": "06:49:23", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3276.7428045035026, "loss": 0.5755896419286728, "entropy": 0.6024290472269058, "sharp": 0.1778944954276085} +{"time": "06:50:14", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3278.459756061006, "loss": 0.5963110364973545, "entropy": 0.5952213108539581, "sharp": 0.18893328122794628} +{"time": "06:51:06", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3280.259019363613, "loss": 0.6257474981248379, "entropy": 0.6058809459209442, "sharp": 0.18065770715475082} +{"time": "06:52:02", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3282.658107966944, "loss": 0.6027997732162476, "entropy": 0.5877076126635075, "sharp": 0.1966692004352808} +{"time": "06:52:58", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3284.523572077551, "loss": 0.5845214314758778, "entropy": 0.5873317420482635, "sharp": 0.1952001266181469} +{"time": "06:53:53", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3286.6416338650647, "loss": 0.5627817288041115, "entropy": 0.5550961457192898, "sharp": 0.23409852385520935} +{"time": "06:54:13", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5731131434440613, "noul_T": 0.8532130995982341, "noul_auc": 0.7512615820830029, "noul_acc": 0.704, "noul_f1": 0.765079365079365, "noul_nll": 0.5736514215767383, "noul_ece": 0.052701245307922376, "noul_brier": 0.1954938496230899, "noul_sharp": 0.17823639512062073, "entropy": 0.5969312787055969} +{"time": "06:54:56", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9113372, "tok_per_s": 3214.371023602123, "rows_per_s": 5.940329960631729, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5595800280570984, "noul_T": 0.8495371650607102, "choice_ce": 0.6842275261878967, "choice_T": 1.0560892190772713, "score_ce": 1.068756341934204, "score_T": 0.960112119036335, "noul_auc": 0.7447338666666667, "noul_acc": 0.695, "noul_f1": 0.7677075399847677, "noul_nll": 0.5717920997068286, "noul_ece": 0.04427964748442174, "noul_brier": 0.19495265463736913, "noul_sharp": 0.1802748143672943, "entropy": 0.5956453680992126}, "dropped_long": 0, "peak_gib": 18.284047603607178} diff --git a/e28/S1-E28-E18A100-rt-syn/DONE b/e28/S1-E28-E18A100-rt-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..af29c858432784548b8f944309a1be2772269c43 --- /dev/null +++ b/e28/S1-E28-E18A100-rt-syn/DONE @@ -0,0 +1 @@ +2026-10-21 05:41:56 \ No newline at end of file diff --git a/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..99aa67ae57369b0e04a5c9046829dafa6045669c --- /dev/null +++ b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "out_proj", + "v_proj", + "q_proj", + "k_proj", + "in_proj_qkv", + "in_proj_z", + "gate_proj", + "o_proj", + "up_proj", + "down_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..06c534fef9aeda2ecca4e33bf3947f7a3e975124 --- /dev/null +++ b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bfe4f81b3bbc62200aedfb4e41c5a27c4008f64325bd8f3696c58e61dea1d9a4 +size 160486456 diff --git a/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..66bee46f4b9b6af0f9b61043a7b012fe5e7bf868 --- /dev/null +++ b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:674988e4c1c9de8f20725c8477e8412211e6b386d03f787a68b365e10e0ecccb +size 176641832 diff --git a/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..57c01499c85b155607bb057116b663cb8dd9cb45 --- /dev/null +++ b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.0003491307725198567, + -0.0003493900876492262, + -0.0028010269161313772, + -0.001637146924622357, + -0.0005709006800316274, + 0.001909777638502419, + 0.0012821187265217304, + -0.002416141564026475, + -0.00029356408049352467, + -0.0006167300161905587, + -0.00018578222079668194, + 0.0005398895591497421, + 0.00042200679308734834, + -0.0036859940737485886, + -0.004552128259092569, + -0.00518615311011672, + -0.005845425650477409, + -0.004403236787766218, + -0.00495922239497304, + -0.005086651537567377, + -0.003894349792972207, + -0.0034869785886257887, + -0.005726017523556948, + -0.0035390020348131657 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-E18A100-rt-syn/ckpt-16000/calibration.json b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..a476836e13bc04a90c001e1db0241eb2f3e674c3 --- /dev/null +++ b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.8950098446145761, + "choice": 0.979539418518294, + "score": 0.9799331682889751 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.8950098446145761, + "ce_before": 0.5527104139328003, + "ce_after": 0.5517128109931946 + }, + "choice": { + "n": 60, + "T": 0.979539418518294, + "ce_before": 0.6627646684646606, + "ce_after": 0.6626709699630737 + }, + "score": { + "n": 53, + "T": 0.9799331682889751, + "ce_before": 1.0634764432907104, + "ce_after": 1.0633518695831299 + } + }, + "source": "models/e28/S1-E28-E18A100-rt-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-E18A100-rt-syn/ckpt-16000/head.safetensors b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ff0a09026ba9fd5e869a73d0af5f970eb465bc52 --- /dev/null +++ b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d934b88b444bb3cdf13f80e6ae05713ce4742c3c68bf0753ad84e6bbef715ffa +size 393464 diff --git a/e28/S1-E28-E18A100-rt-syn/ckpt-16000/judge_config.json b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..e88e5c231dda68e39232c9336cfb1100578ac537 --- /dev/null +++ b/e28/S1-E28-E18A100-rt-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-E18A100-rt-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-E18A100-rt", + "data_manifest": { + "arm": "G-d512-n16k-a100", + "name": null, + "train_sha256": "e68e935b88943ba54ed4296ecedd5f5b36a7b9f114cd4d203c4f67b6ff70b912", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9113372, + "tok_per_s": 3211.5324866181877, + "rows_per_s": 5.935084194122323, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5527104139328003, + "noul_T": 0.8950098446145761, + "choice_ce": 0.6627646684646606, + "choice_T": 0.979539418518294, + "score_ce": 1.0634764432907104, + "score_T": 0.9799331682889751, + "noul_auc": 0.7475370666666667, + "noul_acc": 0.709, + "noul_f1": 0.7826736370425691, + "noul_nll": 0.5664274678044021, + "noul_ece": 0.02627825298905373, + "noul_brier": 0.19344517944405668, + "noul_sharp": 0.18980935215950012, + "entropy": 0.5791373252868652 + }, + "dropped_long": 0, + "peak_gib": 18.284047603607178 + } +} \ No newline at end of file diff --git a/e28/S1-E28-E18A100-rt-syn/train_log.jsonl b/e28/S1-E28-E18A100-rt-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..6805e3848e45e3d38e6f47fe2c755df289bb5977 --- /dev/null +++ b/e28/S1-E28-E18A100-rt-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "04:54:38", "event": "start", "data": "data/e28/E28-E18A100-rt", "data_manifest": {"arm": "G-d512-n16k-a100", "name": null, "train_sha256": "e68e935b88943ba54ed4296ecedd5f5b36a7b9f114cd4d203c4f67b6ff70b912", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.754826784133911} +{"time": "04:55:32", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3324.3171051331515, "loss": 0.5821816921234131, "entropy": 0.5290134195238352, "sharp": 0.2557740639895201} +{"time": "04:56:26", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3342.619469628141, "loss": 0.6144347507506609, "entropy": 0.5189478620886803, "sharp": 0.2612189743667841} +{"time": "04:57:21", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3349.542457682998, "loss": 0.6007676869630814, "entropy": 0.6021645925939083, "sharp": 0.18828725535422564} +{"time": "04:58:14", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3355.5349822500643, "loss": 0.5568909235298634, "entropy": 0.6301036849617958, "sharp": 0.15472890809178352} +{"time": "04:59:08", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3366.9007224387783, "loss": 0.5048067662864923, "entropy": 0.5275742635130882, "sharp": 0.24421783722937107} +{"time": "05:00:01", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3373.9622074492613, "loss": 0.6004810445010662, "entropy": 0.6613579243421555, "sharp": 0.10816137958317995} +{"time": "05:00:56", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3370.7861903597727, "loss": 0.5786934606730938, "entropy": 0.5722468830645084, "sharp": 0.2008360568434} +{"time": "05:01:50", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3371.951818793317, "loss": 0.6301903324201703, "entropy": 0.5992131233215332, "sharp": 0.1787990666925907} +{"time": "05:02:45", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3376.198985771632, "loss": 0.6347234332934022, "entropy": 0.5524705238640308, "sharp": 0.22991735022515059} +{"time": "05:03:39", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3375.2733190336553, "loss": 0.5359861133620143, "entropy": 0.6075579188764095, "sharp": 0.17376595921814442} +{"time": "05:03:59", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.5762508511543274, "noul_T": 0.8315131925978888, "noul_auc": 0.7390255016792964, "noul_acc": 0.698, "noul_f1": 0.7644305772230889, "noul_nll": 0.5822278542816639, "noul_ece": 0.04042269921302796, "noul_brier": 0.19934762721023885, "noul_sharp": 0.17159971594810486, "entropy": 0.6025501489639282} +{"time": "05:04:52", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3270.170828666978, "loss": 0.6028715297579765, "entropy": 0.5942024327814579, "sharp": 0.17740307934582233} +{"time": "05:05:49", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3282.816554544617, "loss": 0.5894755460321903, "entropy": 0.5931592807173729, "sharp": 0.17538204602897167} +{"time": "05:06:45", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3293.4199258042568, "loss": 0.571539293974638, "entropy": 0.5247620716691017, "sharp": 0.2440053764730692} +{"time": "05:07:39", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3299.1396210336193, "loss": 0.5822777207940817, "entropy": 0.5939646065235138, "sharp": 0.190550297498703} +{"time": "05:08:34", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3304.9054761201787, "loss": 0.6522177197039127, "entropy": 0.6408785283565521, "sharp": 0.1347259385511279} +{"time": "05:09:27", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3309.6455791802023, "loss": 0.6231949385255575, "entropy": 0.5634317472577095, "sharp": 0.1964956596493721} +{"time": "05:10:21", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3314.5241506812804, "loss": 0.5992662776261568, "entropy": 0.5705082695931196, "sharp": 0.20299063995480537} +{"time": "05:11:15", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3318.6283858023417, "loss": 0.5902198925614357, "entropy": 0.6148689202964306, "sharp": 0.15836940333247185} +{"time": "05:12:10", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3320.459810493686, "loss": 0.6276428215205669, "entropy": 0.552248153835535, "sharp": 0.21375497709959745} +{"time": "05:13:04", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3324.3459955267745, "loss": 0.5765422154217958, "entropy": 0.5640443600714207, "sharp": 0.21447045169770718} +{"time": "05:13:24", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5789938569068909, "noul_T": 1.0509989837525628, "noul_auc": 0.7417258780442523, "noul_acc": 0.68, "noul_f1": 0.7777777777777778, "noul_nll": 0.5867046617344022, "noul_ece": 0.0768196873664856, "noul_brier": 0.20149740134890642, "noul_sharp": 0.2089177817106247, "entropy": 0.5699007511138916} +{"time": "05:14:19", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3270.421274806653, "loss": 0.5589456427842379, "entropy": 0.580105422064662, "sharp": 0.2016209289431572} +{"time": "05:15:13", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3271.474330897922, "loss": 0.6095458008348942, "entropy": 0.5692577138543129, "sharp": 0.20771372877061367} +{"time": "05:16:06", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3275.132532625415, "loss": 0.5523778740316629, "entropy": 0.5647548958659172, "sharp": 0.22760403528809547} +{"time": "05:16:59", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3278.9549972163577, "loss": 0.5607727654278278, "entropy": 0.5328236278146505, "sharp": 0.23594816960394382} +{"time": "05:17:57", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3282.0149123861243, "loss": 0.6025251597166061, "entropy": 0.5898406133055687, "sharp": 0.18346295692026615} +{"time": "05:18:48", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3284.25290457319, "loss": 0.5806223824620247, "entropy": 0.5832313634455204, "sharp": 0.17723227478563786} +{"time": "05:19:42", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3288.008156761604, "loss": 0.6405539694242179, "entropy": 0.5990578662604094, "sharp": 0.17211627960205078} +{"time": "05:20:35", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3291.6077544568147, "loss": 0.6058562472462654, "entropy": 0.539489358663559, "sharp": 0.2305415105074644} +{"time": "05:21:30", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3294.54182375482, "loss": 0.599469605833292, "entropy": 0.5510324835777283, "sharp": 0.21221148874610662} +{"time": "05:22:25", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3298.2929114311473, "loss": 0.5266063790768385, "entropy": 0.576934652402997, "sharp": 0.17126601748168468} +{"time": "05:22:44", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.57090163230896, "noul_T": 0.7756709156364596, "noul_auc": 0.7441224620681508, "noul_acc": 0.7, "noul_f1": 0.7692307692307693, "noul_nll": 0.5765361273288727, "noul_ece": 0.06336354553699494, "noul_brier": 0.19727589353971983, "noul_sharp": 0.1659199595451355, "entropy": 0.6051366925239563} +{"time": "05:23:37", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3262.4900043871453, "loss": 0.5625473354011774, "entropy": 0.5803492031991482, "sharp": 0.18244618922472} +{"time": "05:24:28", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3265.229639956781, "loss": 0.6385385170578957, "entropy": 0.6035343185067177, "sharp": 0.16779274493455887} +{"time": "05:25:22", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3269.351559759809, "loss": 0.5611035637557507, "entropy": 0.5464083477854729, "sharp": 0.22834619507193565} +{"time": "05:26:15", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3272.9586242924706, "loss": 0.6386581789702177, "entropy": 0.6003948114812374, "sharp": 0.167050426825881} +{"time": "05:27:07", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3275.8133966874448, "loss": 0.6118316054344177, "entropy": 0.5887851566076279, "sharp": 0.17444227263331413} +{"time": "05:28:00", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3279.414812985462, "loss": 0.6040074042975903, "entropy": 0.6173115968704224, "sharp": 0.16369462944567204} +{"time": "05:28:52", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3283.0091197545535, "loss": 0.6168727185577154, "entropy": 0.5541434902697802, "sharp": 0.21848703734576702} +{"time": "05:29:48", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3285.94642100366, "loss": 0.5934649743139744, "entropy": 0.5759682320058346, "sharp": 0.19066925160586834} +{"time": "05:30:43", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3289.937870578109, "loss": 0.626618885435164, "entropy": 0.6006878651678562, "sharp": 0.1617059651762247} +{"time": "05:31:35", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3291.229789227373, "loss": 0.5811722055077553, "entropy": 0.5933747291564941, "sharp": 0.1702809166163206} +{"time": "05:31:55", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5681905746459961, "noul_T": 0.9391060152874654, "noul_auc": 0.7450338390913234, "noul_acc": 0.698, "noul_f1": 0.7722473604826546, "noul_nll": 0.5738915740866214, "noul_ece": 0.035751788198947906, "noul_brier": 0.19638482748407443, "noul_sharp": 0.1894688606262207, "entropy": 0.5781760811805725} +{"time": "05:32:48", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3264.4773228562294, "loss": 0.5989872589707375, "entropy": 0.5823234394192696, "sharp": 0.19112121406942606} +{"time": "05:33:39", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3266.3483477236264, "loss": 0.6516177747398615, "entropy": 0.6142925452440977, "sharp": 0.1551995575428009} +{"time": "05:34:34", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3269.2346552535387, "loss": 0.5770297460258007, "entropy": 0.5725030777975917, "sharp": 0.18772407714277506} +{"time": "05:35:27", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3271.4504029767604, "loss": 0.5983109809458256, "entropy": 0.5960347801446915, "sharp": 0.1682441271841526} +{"time": "05:36:22", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3273.740264407381, "loss": 0.5502499658614397, "entropy": 0.5791618190705776, "sharp": 0.19975292682647705} +{"time": "05:37:14", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3275.4269044173816, "loss": 0.5780162587761879, "entropy": 0.5901758074760437, "sharp": 0.19233804941177368} +{"time": "05:38:06", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3277.5185757402824, "loss": 0.5851897373795509, "entropy": 0.5742927268147469, "sharp": 0.20923052355647087} +{"time": "05:39:02", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3279.550906167545, "loss": 0.5837513078004122, "entropy": 0.5576575566083193, "sharp": 0.22102190181612968} +{"time": "05:39:57", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3281.5783665266927, "loss": 0.6089657805860043, "entropy": 0.5673855915665627, "sharp": 0.20276403427124023} +{"time": "05:40:53", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3283.682019469101, "loss": 0.5507879704236984, "entropy": 0.5276962127536535, "sharp": 0.24716411530971527} +{"time": "05:41:13", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5671284198760986, "noul_T": 0.915414303765002, "noul_auc": 0.7490337715819142, "noul_acc": 0.71, "noul_f1": 0.7744945567651633, "noul_nll": 0.5708517383821309, "noul_ece": 0.037796863555908215, "noul_brier": 0.19516738459980418, "noul_sharp": 0.1869892030954361, "entropy": 0.5807954668998718} +{"time": "05:41:56", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9113372, "tok_per_s": 3211.5324866181877, "rows_per_s": 5.935084194122323, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5527104139328003, "noul_T": 0.8950098446145761, "choice_ce": 0.6627646684646606, "choice_T": 0.979539418518294, "score_ce": 1.0634764432907104, "score_T": 0.9799331682889751, "noul_auc": 0.7475370666666667, "noul_acc": 0.709, "noul_f1": 0.7826736370425691, "noul_nll": 0.5664274678044021, "noul_ece": 0.02627825298905373, "noul_brier": 0.19344517944405668, "noul_sharp": 0.18980935215950012, "entropy": 0.5791373252868652}, "dropped_long": 0, "peak_gib": 18.284047603607178} diff --git a/e28/S1-E28-E30-a1-rw-syn/DONE b/e28/S1-E28-E30-a1-rw-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..076631f15d5b4dd814653b80dc4244020daf556e --- /dev/null +++ b/e28/S1-E28-E30-a1-rw-syn/DONE @@ -0,0 +1 @@ +2026-10-21 18:40:28 \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..e5f1736af8b8efdc0081468df29206bd01fb47d1 --- /dev/null +++ b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "out_proj", + "v_proj", + "gate_proj", + "in_proj_z", + "up_proj", + "o_proj", + "q_proj", + "k_proj", + "down_proj", + "in_proj_qkv" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..823ed91c8ace7acda06edec18c7f9312250ec9fd --- /dev/null +++ b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d78eef6a2ecf5d7a7fa2e3e767d90b6a47985a8b9f58e1394629c81783d1bcc5 +size 160486456 diff --git a/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8f6b518794ce59ce12a4acfd794115384014dec3 --- /dev/null +++ b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f90749a62d8e87daafce65f9bd5e2326e6c48bd746b4755d9e51db4bb4547bc +size 176641832 diff --git a/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..68d021eb49b4cd01ebf25f6ea8c32984819a8066 --- /dev/null +++ b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.0005390653386712074, + -0.0005393188330344856, + -0.002505986951291561, + -0.0016197734512388706, + -0.0006678372155874968, + 0.0021317743230611086, + 0.001403430593200028, + -0.002808196237310767, + -0.0001560490927658975, + -0.0002977506082970649, + 0.0003498705627862364, + 0.0007481754291802645, + -0.0015342392725870013, + -0.0035954874474555254, + -0.004323368426412344, + -0.0050967843271791935, + -0.005942096468061209, + -0.004360170569270849, + -0.0049414122477173805, + -0.00499636260792613, + -0.003998260945081711, + -0.00342180160805583, + -0.005643993150442839, + -0.0035076222848147154 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/calibration.json b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..b14918aedefffc3a55fac237240e9217de2361cb --- /dev/null +++ b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.7893042861110305, + "choice": 1.1089171735549759, + "score": 0.9725362941237807 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.7893042861110305, + "ce_before": 0.5852800011634827, + "ce_after": 0.5815325975418091 + }, + "choice": { + "n": 60, + "T": 1.1089171735549759, + "ce_before": 0.6824166774749756, + "ce_after": 0.6801118850708008 + }, + "score": { + "n": 53, + "T": 0.9725362941237807, + "ce_before": 1.0658724308013916, + "ce_after": 1.065628170967102 + } + }, + "source": "models/e28/S1-E28-E30-a1-rw-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/head.safetensors b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d4c36a197b919629eceac1d8b71993dc0d0ad56e --- /dev/null +++ b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e188d1d84a9fa6d952cdd6536db406217276ece45e56249448536c47c9d523fd +size 393464 diff --git a/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/judge_config.json b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..394e2e6eca49f1a86c87ab9695f29168103e983f --- /dev/null +++ b/e28/S1-E28-E30-a1-rw-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-E30-a1-rw-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-E30-a1-rw", + "data_manifest": { + "arm": "E28-E30-a1-rw", + "name": null, + "train_sha256": "d26b35d46e3927d8807524aa9b0efcfcb8505a3ac3b2c30a9a2933198ec3e62d", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 8988859, + "tok_per_s": 3275.2473404806465, + "rows_per_s": 6.13667604304651, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5852800011634827, + "noul_T": 0.7893042861110305, + "choice_ce": 0.6824166774749756, + "choice_T": 1.1089171735549759, + "score_ce": 1.0658724308013916, + "score_T": 0.9725362941237807, + "noul_auc": 0.7184704, + "noul_acc": 0.696, + "noul_f1": 0.76793893129771, + "noul_nll": 0.5946063060313463, + "noul_ece": 0.059608851045370115, + "noul_brier": 0.20349168514643015, + "noul_sharp": 0.15316306054592133, + "entropy": 0.6263568997383118 + }, + "dropped_long": 0, + "peak_gib": 18.282336711883545 + } +} \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-rw-syn/train_log.jsonl b/e28/S1-E28-E30-a1-rw-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..b88b41470ddbe74f5bb9ef0a0f9f959d9b28dfcf --- /dev/null +++ b/e28/S1-E28-E30-a1-rw-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "17:54:43", "event": "start", "data": "data/e28/E28-E30-a1-rw", "data_manifest": {"arm": "E28-E30-a1-rw", "name": null, "train_sha256": "d26b35d46e3927d8807524aa9b0efcfcb8505a3ac3b2c30a9a2933198ec3e62d", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.869741678237915} +{"time": "17:55:35", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3377.2851167766844, "loss": 0.6338330507278442, "entropy": 0.6497155390679836, "sharp": 0.12389896251261234} +{"time": "17:56:26", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3412.0501605520712, "loss": 0.6315236687660217, "entropy": 0.645004142075777, "sharp": 0.1294181663542986} +{"time": "17:57:18", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3426.6409228336306, "loss": 0.7279861196875572, "entropy": 0.6225987821817398, "sharp": 0.14857110660523176} +{"time": "17:58:11", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3432.445975142808, "loss": 0.6418519904837012, "entropy": 0.6383919306099415, "sharp": 0.15334222465753555} +{"time": "17:59:01", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3446.264373239778, "loss": 0.6650119200348854, "entropy": 0.669834554195404, "sharp": 0.08912533335387707} +{"time": "17:59:53", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3449.69168789776, "loss": 0.6345913931727409, "entropy": 0.661293052136898, "sharp": 0.10877801477909088} +{"time": "18:00:46", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3447.4641306971735, "loss": 0.6558307074010372, "entropy": 0.6163828521966934, "sharp": 0.17564489413052797} +{"time": "18:01:38", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3446.495126628318, "loss": 0.6458177231252193, "entropy": 0.6574584618210793, "sharp": 0.09814455173909664} +{"time": "18:02:30", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3447.1536995850843, "loss": 0.7036329247057438, "entropy": 0.6312954239547253, "sharp": 0.13076926954090595} +{"time": "18:03:21", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3443.790074357248, "loss": 0.6426000259816647, "entropy": 0.6454283036291599, "sharp": 0.12996309995651245} +{"time": "18:03:41", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.6357130408287048, "noul_T": 0.6211001554924073, "noul_auc": 0.7169161701912204, "noul_acc": 0.67, "noul_f1": 0.7169811320754716, "noul_nll": 0.6331049682497978, "noul_ece": 0.12251776278018953, "noul_brier": 0.22072263777458637, "noul_sharp": 0.0979577973484993, "entropy": 0.66475510597229} +{"time": "18:04:33", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3329.341525297175, "loss": 0.5837785638868809, "entropy": 0.594234935939312, "sharp": 0.17981287464499474} +{"time": "18:05:24", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3340.8014945412456, "loss": 0.6435245983302593, "entropy": 0.6653639450669289, "sharp": 0.08566387463361025} +{"time": "18:06:16", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3349.8566168747334, "loss": 0.5971063524484634, "entropy": 0.6157004665583372, "sharp": 0.1654625367373228} +{"time": "18:07:07", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3355.556275132598, "loss": 0.6818988248705864, "entropy": 0.630051501095295, "sharp": 0.1473753023892641} +{"time": "18:07:59", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3360.7227262118395, "loss": 0.5924976449459791, "entropy": 0.6424582377076149, "sharp": 0.13546935934573412} +{"time": "18:08:49", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3365.882176538322, "loss": 0.7351751737296581, "entropy": 0.6749421320855618, "sharp": 0.08052086178213358} +{"time": "18:09:42", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3372.250622334071, "loss": 0.627573162317276, "entropy": 0.6558959558606148, "sharp": 0.12126355338841677} +{"time": "18:10:34", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3374.350063685714, "loss": 0.5482108443975449, "entropy": 0.6002820134162903, "sharp": 0.15795409120619297} +{"time": "18:11:26", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3378.7029186270856, "loss": 0.6277678199112415, "entropy": 0.6315691769123077, "sharp": 0.1291743665933609} +{"time": "18:12:20", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3386.2736576786465, "loss": 0.5901587307453156, "entropy": 0.640953928232193, "sharp": 0.12728234101086855} +{"time": "18:12:40", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.6395695209503174, "noul_T": 0.5691573306075144, "noul_auc": 0.7214393006025215, "noul_acc": 0.658, "noul_f1": 0.7005253940455342, "noul_nll": 0.6360012717843055, "noul_ece": 0.14528395158052446, "noul_brier": 0.2222243668546396, "noul_sharp": 0.08701810240745544, "entropy": 0.6687576770782471} +{"time": "18:13:32", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3328.7936073683777, "loss": 0.6967102643102407, "entropy": 0.6456865444779396, "sharp": 0.11548532731831074} +{"time": "18:14:23", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3333.543467474346, "loss": 0.6308403629809618, "entropy": 0.6531947795301676, "sharp": 0.10847062431275845} +{"time": "18:15:16", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3340.0488545184285, "loss": 0.6255427449941635, "entropy": 0.6058772392570972, "sharp": 0.16990910656750202} +{"time": "18:16:10", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3344.300160322416, "loss": 0.71213773265481, "entropy": 0.6475676037371159, "sharp": 0.12428359594196081} +{"time": "18:17:00", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3348.6065993451734, "loss": 0.6979744881391525, "entropy": 0.6062525287270546, "sharp": 0.18239591270685196} +{"time": "18:17:53", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3352.879972166113, "loss": 0.5990705713629723, "entropy": 0.6486502215266228, "sharp": 0.1148771271109581} +{"time": "18:18:44", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3358.1137566191123, "loss": 0.6887011472135782, "entropy": 0.6524376422166824, "sharp": 0.1245900746434927} +{"time": "18:19:38", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3363.13589939421, "loss": 0.6507637053728104, "entropy": 0.6164841204881668, "sharp": 0.16536646895110607} +{"time": "18:20:30", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3366.5677356122173, "loss": 0.6710220128297806, "entropy": 0.6520177498459816, "sharp": 0.11389991734176874} +{"time": "18:21:22", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3368.685068466974, "loss": 0.6416098596528172, "entropy": 0.6477503851056099, "sharp": 0.12916356604546309} +{"time": "18:21:42", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.6175177693367004, "noul_T": 0.7215910214407419, "noul_auc": 0.7286628073787784, "noul_acc": 0.674, "noul_f1": 0.7165217391304348, "noul_nll": 0.6141112993955612, "noul_ece": 0.11070850026607516, "noul_brier": 0.21205119923383237, "noul_sharp": 0.12754103541374207, "entropy": 0.6471936702728271} +{"time": "18:22:35", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3331.3107669826086, "loss": 0.6511296331882477, "entropy": 0.6407931558787823, "sharp": 0.13699984084814787} +{"time": "18:23:27", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3334.268322375747, "loss": 0.6851719468832016, "entropy": 0.6582037210464478, "sharp": 0.11040528025478125} +{"time": "18:24:20", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3338.41235718141, "loss": 0.6316585764288902, "entropy": 0.6686232909560204, "sharp": 0.09014521725475788} +{"time": "18:25:12", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3341.2341405458083, "loss": 0.6132487095892429, "entropy": 0.5950825084000826, "sharp": 0.18371212854981422} +{"time": "18:26:04", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3344.3896086410027, "loss": 0.669573157094419, "entropy": 0.6661554910242558, "sharp": 0.09097274485975504} +{"time": "18:26:56", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3348.0805313321516, "loss": 0.6744805574417114, "entropy": 0.6449346467852592, "sharp": 0.12108333129435778} +{"time": "18:27:49", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3350.44983143245, "loss": 0.6190249249339104, "entropy": 0.5869340002536774, "sharp": 0.19311721064150333} +{"time": "18:28:42", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3352.2686568872564, "loss": 0.6518426295369864, "entropy": 0.6288196444511414, "sharp": 0.13674758467823267} +{"time": "18:29:33", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3355.277724566154, "loss": 0.697613812983036, "entropy": 0.637400783598423, "sharp": 0.12889212369918823} +{"time": "18:30:25", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3356.9764095931605, "loss": 0.61271165125072, "entropy": 0.6158593855798244, "sharp": 0.1538755875080824} +{"time": "18:30:45", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.6081778407096863, "noul_T": 0.7508601228503166, "noul_auc": 0.7353631162343252, "noul_acc": 0.706, "noul_f1": 0.7578253706754531, "noul_nll": 0.6025428766608238, "noul_ece": 0.10199650961160661, "noul_brier": 0.20667334150308625, "noul_sharp": 0.13835392892360687, "entropy": 0.63767409324646} +{"time": "18:31:38", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3329.4294478765714, "loss": 0.6279964409768581, "entropy": 0.6595849394798279, "sharp": 0.0933420555666089} +{"time": "18:32:30", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3332.267656872563, "loss": 0.6110328286886215, "entropy": 0.6364967674016953, "sharp": 0.13975976780056953} +{"time": "18:33:21", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3334.403992239488, "loss": 0.6370786568149924, "entropy": 0.6099487831816077, "sharp": 0.15920150559395552} +{"time": "18:34:15", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3338.3739954868947, "loss": 0.6492991335690022, "entropy": 0.6316069588065147, "sharp": 0.13702389132231474} +{"time": "18:35:07", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3340.5295980148544, "loss": 0.5848479829728603, "entropy": 0.6442913822829723, "sharp": 0.11933630052953959} +{"time": "18:35:58", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3342.722650089894, "loss": 0.6680366033688188, "entropy": 0.6582284085452557, "sharp": 0.10449005477130413} +{"time": "18:36:50", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3344.5248878335597, "loss": 0.6666250005364418, "entropy": 0.6220298185944557, "sharp": 0.15243109688162804} +{"time": "18:37:42", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3346.176169703262, "loss": 0.6240421757102013, "entropy": 0.6295576170086861, "sharp": 0.13606482464820147} +{"time": "18:38:33", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3349.000266115592, "loss": 0.6278671249747276, "entropy": 0.6624680310487747, "sharp": 0.09166452661156654} +{"time": "18:39:25", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3351.4645199172123, "loss": 0.7027888223528862, "entropy": 0.6221847385168076, "sharp": 0.15409976709634066} +{"time": "18:39:45", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.6007530093193054, "noul_T": 0.7747629570244622, "noul_auc": 0.7332196924946415, "noul_acc": 0.708, "noul_f1": 0.7660256410256411, "noul_nll": 0.5951983838230371, "noul_ece": 0.08512620198726653, "noul_brier": 0.20347204683407666, "noul_sharp": 0.14822502434253693, "entropy": 0.6300143599510193} +{"time": "18:40:28", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 8988859, "tok_per_s": 3275.2473404806465, "rows_per_s": 6.13667604304651, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5852800011634827, "noul_T": 0.7893042861110305, "choice_ce": 0.6824166774749756, "choice_T": 1.1089171735549759, "score_ce": 1.0658724308013916, "score_T": 0.9725362941237807, "noul_auc": 0.7184704, "noul_acc": 0.696, "noul_f1": 0.76793893129771, "noul_nll": 0.5946063060313463, "noul_ece": 0.059608851045370115, "noul_brier": 0.20349168514643015, "noul_sharp": 0.15316306054592133, "entropy": 0.6263568997383118}, "dropped_long": 0, "peak_gib": 18.282336711883545} diff --git a/e28/S1-E28-E30-a1-uni-syn/DONE b/e28/S1-E28-E30-a1-uni-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..e46354ac75be7d68bd73cc9624c5b07fd2b49e2d --- /dev/null +++ b/e28/S1-E28-E30-a1-uni-syn/DONE @@ -0,0 +1 @@ +2026-10-21 17:27:13 \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..3afa68fb815082a1d8f8bdd9741f0a4879cca094 --- /dev/null +++ b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "out_proj", + "in_proj_z", + "k_proj", + "o_proj", + "v_proj", + "gate_proj", + "in_proj_qkv", + "down_proj", + "q_proj", + "up_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e6d69186f3c6d785800f4616b34513b38ad2f05b --- /dev/null +++ b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d0bc376b8094b519732e128bc6c2e77088496bc2b7ef13135e140ee0912c34be +size 160486456 diff --git a/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..440eea1c27a005e4856bd068723aeefe88e87985 --- /dev/null +++ b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cd954b72350ecb7f709fe16b769604f928bdb583b488ca6aa79d7f8adf78ca5b +size 176641832 diff --git a/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..831d785ed5ff5735c4edf38bec07e6a4b27e394d --- /dev/null +++ b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.0006060886080376804, + -0.0006063475739210844, + -0.0024707692209631205, + -0.0015818045940250158, + -0.0006252293824218214, + 0.002087934873998165, + 0.0011421249946579337, + -0.0028095112647861242, + -0.00038323772605508566, + -0.0003126315423287451, + -3.52837705577258e-05, + 0.0008903947309590876, + -0.0002762395015452057, + -0.0034320163540542126, + -0.0041916826739907265, + -0.005088407546281815, + -0.005943010561168194, + -0.0043299696408212185, + -0.004925391636788845, + -0.004935245495289564, + -0.0038020077627152205, + -0.0033371136523783207, + -0.005604666657745838, + -0.0034593206364661455 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/calibration.json b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..c2e83ade4bc71de26301074381486db505b8ad51 --- /dev/null +++ b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.8655641338951023, + "choice": 1.1227301938835272, + "score": 0.9718134647327761 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.8655641338951023, + "ce_before": 0.5874700546264648, + "ce_after": 0.586046576499939 + }, + "choice": { + "n": 60, + "T": 1.1227301938835272, + "ce_before": 0.6819011569023132, + "ce_after": 0.6787732839584351 + }, + "score": { + "n": 53, + "T": 0.9718134647327761, + "ce_before": 1.0681289434432983, + "ce_after": 1.067886471748352 + } + }, + "source": "models/e28/S1-E28-E30-a1-uni-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/head.safetensors b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c8c08134a1668351f08ef24bd177173c7562f6a6 --- /dev/null +++ b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77e7ef36ad6ee4b9aa196e486c3dbd50319e462a6a88e0df183135a69056539e +size 393464 diff --git a/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/judge_config.json b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..adfd086d7400369968ab38a948ff01aadbe395d6 --- /dev/null +++ b/e28/S1-E28-E30-a1-uni-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-E30-a1-uni-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-E30-a1-uni", + "data_manifest": { + "arm": "E28-E30-a1-uni", + "name": null, + "train_sha256": "12a197d7649dce14980e9b45d1a775704a077b314dd01d1eb1ccf6ad58e6b493", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 8947586, + "tok_per_s": 3272.6624281029794, + "rows_per_s": 6.1601174412470865, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5874700546264648, + "noul_T": 0.8655641338951023, + "choice_ce": 0.6819011569023132, + "choice_T": 1.1227301938835272, + "score_ce": 1.0681289434432983, + "score_T": 0.9718134647327761, + "noul_auc": 0.7201664, + "noul_acc": 0.688, + "noul_f1": 0.7554858934169278, + "noul_nll": 0.5952357437387109, + "noul_ece": 0.06844655176997184, + "noul_brier": 0.20430087828807472, + "noul_sharp": 0.1595011055469513, + "entropy": 0.6177899837493896 + }, + "dropped_long": 0, + "peak_gib": 18.282336235046387 + } +} \ No newline at end of file diff --git a/e28/S1-E28-E30-a1-uni-syn/train_log.jsonl b/e28/S1-E28-E30-a1-uni-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..c8aca8827cb6bf83648c325e90c44f60af384edc --- /dev/null +++ b/e28/S1-E28-E30-a1-uni-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "16:41:38", "event": "start", "data": "data/e28/E28-E30-a1-uni", "data_manifest": {"arm": "E28-E30-a1-uni", "name": null, "train_sha256": "12a197d7649dce14980e9b45d1a775704a077b314dd01d1eb1ccf6ad58e6b493", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.735321521759033} +{"time": "16:42:30", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3409.304770646953, "loss": 0.6041513346135616, "entropy": 0.5120556354522705, "sharp": 0.27248259633779526} +{"time": "16:43:21", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3421.4667461606596, "loss": 0.556976668536663, "entropy": 0.5864505395293236, "sharp": 0.20490439236164093} +{"time": "16:44:13", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3432.8491953377434, "loss": 0.5619225632399321, "entropy": 0.5766297020018101, "sharp": 0.21862560790032148} +{"time": "16:45:04", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3424.6372606763744, "loss": 0.609934551641345, "entropy": 0.6585803888738155, "sharp": 0.09875787049531937} +{"time": "16:45:55", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3430.6862666371885, "loss": 0.6419296786189079, "entropy": 0.6163046360015869, "sharp": 0.1718506794422865} +{"time": "16:46:47", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3434.4303609877184, "loss": 0.5647634267807007, "entropy": 0.5451114177703857, "sharp": 0.25240151211619377} +{"time": "16:47:37", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3440.031452329613, "loss": 0.5020686089992523, "entropy": 0.5977845713496208, "sharp": 0.20244209468364716} +{"time": "16:48:30", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3440.302208558174, "loss": 0.5558787388727069, "entropy": 0.599686473608017, "sharp": 0.18862658645957708} +{"time": "16:49:23", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3449.2628462745943, "loss": 0.634467352181673, "entropy": 0.5588158629834652, "sharp": 0.23588495515286922} +{"time": "16:50:16", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3453.6128459218544, "loss": 0.5926288142800331, "entropy": 0.5601833313703537, "sharp": 0.2304825223982334} +{"time": "16:50:36", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.6498395800590515, "noul_T": 1.349077282178232, "noul_auc": 0.7146883596901319, "noul_acc": 0.632, "noul_f1": 0.6617647058823529, "noul_nll": 0.6422284835875034, "noul_ece": 0.13594646209478378, "noul_brier": 0.22519311402935216, "noul_sharp": 0.17081335186958313, "entropy": 0.6116818189620972} +{"time": "16:51:28", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3336.697677767694, "loss": 0.6386707499623299, "entropy": 0.6384924352169037, "sharp": 0.13602092117071152} +{"time": "16:52:21", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3346.2196664894504, "loss": 0.560898169875145, "entropy": 0.5595465749502182, "sharp": 0.23700227215886116} +{"time": "16:53:14", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3353.9533567875774, "loss": 0.511741291731596, "entropy": 0.5179191660135984, "sharp": 0.26211291924118996} +{"time": "16:54:06", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3362.345738936209, "loss": 0.5394777879118919, "entropy": 0.5748411938548088, "sharp": 0.20575784519314766} +{"time": "16:54:58", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3367.160286999917, "loss": 0.5847659632563591, "entropy": 0.6158844381570816, "sharp": 0.15531163476407528} +{"time": "16:55:49", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3370.7225356064714, "loss": 0.6675980417057872, "entropy": 0.523243511095643, "sharp": 0.2477388009428978} +{"time": "16:56:42", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3377.63274379463, "loss": 0.6172627545893192, "entropy": 0.5247740596532822, "sharp": 0.2580416016280651} +{"time": "16:57:35", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3383.3926327576332, "loss": 0.5857275687158108, "entropy": 0.6146125830709934, "sharp": 0.15807517245411873} +{"time": "16:58:28", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3388.1870808955923, "loss": 0.5469414964318275, "entropy": 0.5764984805136919, "sharp": 0.20309966430068016} +{"time": "16:59:20", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3391.568360485484, "loss": 0.5608526840806007, "entropy": 0.539695255458355, "sharp": 0.22885571420192719} +{"time": "16:59:40", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.6151646375656128, "noul_T": 1.0223011331649534, "noul_auc": 0.7449156976253566, "noul_acc": 0.712, "noul_f1": 0.7575757575757576, "noul_nll": 0.6018782912343741, "noul_ece": 0.10552632319927217, "noul_brier": 0.20531924337192112, "noul_sharp": 0.16664859652519226, "entropy": 0.6123971939086914} +{"time": "17:00:32", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3336.157434362825, "loss": 0.6180928945541382, "entropy": 0.5792511850595474, "sharp": 0.20438549295067787} +{"time": "17:01:22", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3338.5330154686594, "loss": 0.6049556247889996, "entropy": 0.5932673662900925, "sharp": 0.1860632235184312} +{"time": "17:02:14", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3342.616207086156, "loss": 0.5297068953514099, "entropy": 0.5577865913510323, "sharp": 0.23085156828165054} +{"time": "17:03:05", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3347.3138305078432, "loss": 0.5956302136182785, "entropy": 0.6197281256318092, "sharp": 0.15559537149965763} +{"time": "17:03:57", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3350.9253249050444, "loss": 0.6904696580022573, "entropy": 0.615265004336834, "sharp": 0.16924737952649593} +{"time": "17:04:49", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3356.155554477883, "loss": 0.5945560857653618, "entropy": 0.5844353884458542, "sharp": 0.18966596201062202} +{"time": "17:05:42", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3359.2732168530465, "loss": 0.5674677193164825, "entropy": 0.6142416372895241, "sharp": 0.1503648804500699} +{"time": "17:06:34", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3362.4347313497724, "loss": 0.6052850186824799, "entropy": 0.5131549052894115, "sharp": 0.26372173614799976} +{"time": "17:07:27", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3366.5519811913123, "loss": 0.5395555719733238, "entropy": 0.5875923112034798, "sharp": 0.201644416898489} +{"time": "17:08:19", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3368.63175715889, "loss": 0.5460652261972427, "entropy": 0.5314070209860802, "sharp": 0.25500188767910004} +{"time": "17:08:39", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.6281391978263855, "noul_T": 0.9941926310268644, "noul_auc": 0.7312956743346104, "noul_acc": 0.646, "noul_f1": 0.6810810810810811, "noul_nll": 0.617037369325757, "noul_ece": 0.11867893391847614, "noul_brier": 0.214179000216506, "noul_sharp": 0.14709702134132385, "entropy": 0.628792405128479} +{"time": "17:09:31", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3330.6768389156264, "loss": 0.6375809237360954, "entropy": 0.5859684571623802, "sharp": 0.1949059907346964} +{"time": "17:10:24", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3334.8513496572873, "loss": 0.5917091816663742, "entropy": 0.6200159788131714, "sharp": 0.17136229388415813} +{"time": "17:11:14", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3338.3167659013816, "loss": 0.6080724149942398, "entropy": 0.5728140249848366, "sharp": 0.22468452528119087} +{"time": "17:12:03", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3340.3203394983198, "loss": 0.5881139039993286, "entropy": 0.594871498644352, "sharp": 0.18038641288876534} +{"time": "17:12:55", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3342.5934205986796, "loss": 0.5185284893959761, "entropy": 0.5448028296232224, "sharp": 0.22866988740861416} +{"time": "17:13:47", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3345.9009569891173, "loss": 0.5583254434168339, "entropy": 0.5857109576463699, "sharp": 0.19378767907619476} +{"time": "17:14:39", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3348.2990593702725, "loss": 0.5837229005992413, "entropy": 0.5581596493721008, "sharp": 0.21905477344989777} +{"time": "17:15:31", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3350.080792768038, "loss": 0.5410677660256624, "entropy": 0.49715545773506165, "sharp": 0.27314646169543266} +{"time": "17:16:21", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3353.155419572864, "loss": 0.718022670596838, "entropy": 0.5609892904758453, "sharp": 0.22347460687160492} +{"time": "17:17:12", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3355.120472988009, "loss": 0.5695267729461193, "entropy": 0.598109669983387, "sharp": 0.17371178418397903} +{"time": "17:17:32", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.6086302399635315, "noul_T": 0.828350931287299, "noul_auc": 0.7362407385529357, "noul_acc": 0.688, "noul_f1": 0.7417218543046358, "noul_nll": 0.5997099155783653, "noul_ece": 0.0929402739405632, "noul_brier": 0.20585448037746093, "noul_sharp": 0.1464211344718933, "entropy": 0.6296308636665344} +{"time": "17:18:23", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3327.2761251380134, "loss": 0.6214786693453789, "entropy": 0.5436572656035423, "sharp": 0.23656857758760452} +{"time": "17:19:15", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3329.441187386622, "loss": 0.639856792986393, "entropy": 0.5497752204537392, "sharp": 0.242206497117877} +{"time": "17:20:07", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3332.1759102448664, "loss": 0.570521280169487, "entropy": 0.5385730378329754, "sharp": 0.23419552110135555} +{"time": "17:20:59", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3333.950008187398, "loss": 0.5842433180660009, "entropy": 0.5844602901488543, "sharp": 0.19395216926932335} +{"time": "17:21:51", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3335.7967368606987, "loss": 0.4905189909040928, "entropy": 0.5526576116681099, "sharp": 0.2242641467601061} +{"time": "17:22:42", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3338.5657664491005, "loss": 0.6897539012134075, "entropy": 0.592585101723671, "sharp": 0.17433451674878597} +{"time": "17:23:35", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3341.8162467012125, "loss": 0.5974264889955521, "entropy": 0.5931247770786285, "sharp": 0.18097318522632122} +{"time": "17:24:26", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3344.4980719493074, "loss": 0.6506990473717451, "entropy": 0.5466227047145367, "sharp": 0.24038755428045988} +{"time": "17:25:17", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3346.477627106757, "loss": 0.5294239781796932, "entropy": 0.5471327640116215, "sharp": 0.22505945153534412} +{"time": "17:26:10", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3349.0037538328743, "loss": 0.5496084466576576, "entropy": 0.5368413776159286, "sharp": 0.2411971464753151} +{"time": "17:26:30", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.604293942451477, "noul_T": 0.8665043970684547, "noul_auc": 0.7400043881115931, "noul_acc": 0.7, "noul_f1": 0.7524752475247525, "noul_nll": 0.5948870942890644, "noul_ece": 0.09654653149843216, "noul_brier": 0.203694097850999, "noul_sharp": 0.15578387677669525, "entropy": 0.6217830181121826} +{"time": "17:27:13", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 8947586, "tok_per_s": 3272.6624281029794, "rows_per_s": 6.1601174412470865, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5874700546264648, "noul_T": 0.8655641338951023, "choice_ce": 0.6819011569023132, "choice_T": 1.1227301938835272, "score_ce": 1.0681289434432983, "score_T": 0.9718134647327761, "noul_auc": 0.7201664, "noul_acc": 0.688, "noul_f1": 0.7554858934169278, "noul_nll": 0.5952357437387109, "noul_ece": 0.06844655176997184, "noul_brier": 0.20430087828807472, "noul_sharp": 0.1595011055469513, "entropy": 0.6177899837493896}, "dropped_long": 0, "peak_gib": 18.282336235046387} diff --git a/e28/S1-E28-E30-b-acc-syn/DONE b/e28/S1-E28-E30-b-acc-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..31e4777911d7a525b1788daafc1f9a5a84aa759a --- /dev/null +++ b/e28/S1-E28-E30-b-acc-syn/DONE @@ -0,0 +1 @@ +2026-10-21 10:42:01 \ No newline at end of file diff --git a/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..3231bf0097507f93c206f1bed58fdf4c9d445b31 --- /dev/null +++ b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "v_proj", + "in_proj_z", + "out_proj", + "k_proj", + "down_proj", + "up_proj", + "q_proj", + "in_proj_qkv", + "o_proj", + "gate_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5544623376e1b4d5683fb098fb21617fe982edde --- /dev/null +++ b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b19107b4b258becac076df6be4f6d8f1dd2363059d7872d5e03b5e3edac8a10 +size 160486456 diff --git a/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..621d0b53299eae750a523c4e909c10263cfa6991 --- /dev/null +++ b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4a390b414b1ae09d58792786ccc67ff6a1fa4a29c4ce5ab74aefb194b79d0219 +size 176641832 diff --git a/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..1569f22caa12413da18b7682b1f41287403d3001 --- /dev/null +++ b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.0003408764023333788, + -0.00034113190486095846, + -0.002685137093067169, + -0.0019023797940462828, + -0.0005612417589873075, + 0.0018154303543269634, + 0.0007902936195023358, + -0.0033744939137250185, + -0.0007680312264710665, + -0.00043316202936694026, + -5.294220682117157e-05, + 0.00034113373840227723, + -1.9338880520081148e-05, + -0.0038968417793512344, + -0.004540482070297003, + -0.005175251513719559, + -0.006013345438987017, + -0.0044153304770588875, + -0.004955203738063574, + -0.0050728293135762215, + -0.004119143355637789, + -0.003500294405966997, + -0.005672055762261152, + -0.0035358895547688007 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-acc-syn/ckpt-16000/calibration.json b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..d5f771154c2bb61da922a0f1a5fc0da1614482ae --- /dev/null +++ b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.902481163861105, + "choice": 1.0356540084892951, + "score": 0.9725893429803429 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.902481163861105, + "ce_before": 0.5696535110473633, + "ce_after": 0.568800151348114 + }, + "choice": { + "n": 60, + "T": 1.0356540084892951, + "ce_before": 0.6680582165718079, + "ce_after": 0.6677979230880737 + }, + "score": { + "n": 53, + "T": 0.9725893429803429, + "ce_before": 1.0546555519104004, + "ce_after": 1.054425597190857 + } + }, + "source": "models/e28/S1-E28-E30-b-acc-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-acc-syn/ckpt-16000/head.safetensors b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2b957bb41c493fe6606220a474ee1185ec6b461c --- /dev/null +++ b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:830b5175cbf0224380d68a74be4296733aca26721256f84defa624fcff5fa050 +size 393464 diff --git a/e28/S1-E28-E30-b-acc-syn/ckpt-16000/judge_config.json b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..90716893cf4ea1282bc7a6fac75b0cb30aac005a --- /dev/null +++ b/e28/S1-E28-E30-b-acc-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-E30-b-acc-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-E30-b-acc", + "data_manifest": { + "arm": "E28-E30-b-acc", + "name": null, + "train_sha256": "14edf2d2782d5bfda001c5df6a3ab99ffc94c0d4a6fbd20989f471757b9e06fe", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9813814, + "tok_per_s": 3313.573213319746, + "rows_per_s": 5.686596466348972, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5696535110473633, + "noul_T": 0.902481163861105, + "choice_ce": 0.6680582165718079, + "choice_T": 1.0356540084892951, + "score_ce": 1.0546555519104004, + "score_T": 0.9725893429803429, + "noul_auc": 0.7243605333333334, + "noul_acc": 0.703, + "noul_f1": 0.7889125799573561, + "noul_nll": 0.5870741973295808, + "noul_ece": 0.04549183717370033, + "noul_brier": 0.20079337328284533, + "noul_sharp": 0.18132315576076508, + "entropy": 0.5971601605415344 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-acc-syn/train_log.jsonl b/e28/S1-E28-E30-b-acc-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..e35d4820aa5892fafc2f7a3b76cb99ccac8a5b3a --- /dev/null +++ b/e28/S1-E28-E30-b-acc-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "09:52:39", "event": "start", "data": "data/e28/E28-E30-b-acc", "data_manifest": {"arm": "E28-E30-b-acc", "name": null, "train_sha256": "14edf2d2782d5bfda001c5df6a3ab99ffc94c0d4a6fbd20989f471757b9e06fe", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.783592700958252} +{"time": "09:53:34", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3433.748158528713, "loss": 0.6055417880415916, "entropy": 0.6421226412057877, "sharp": 0.13293416891247034} +{"time": "09:54:31", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3452.4359347571435, "loss": 0.6198242194950581, "entropy": 0.6231202259659767, "sharp": 0.15255100652575493} +{"time": "09:55:26", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3459.1338993137742, "loss": 0.6869230140000582, "entropy": 0.6299545131623745, "sharp": 0.15643970482051373} +{"time": "09:56:22", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3463.627189645415, "loss": 0.6159832216799259, "entropy": 0.6474946811795235, "sharp": 0.12117885425686836} +{"time": "09:57:19", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3462.8621068900675, "loss": 0.6545930206775665, "entropy": 0.6262377500534058, "sharp": 0.16176053509116173} +{"time": "09:58:14", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3462.6815304934767, "loss": 0.6711318790912628, "entropy": 0.5956233609467745, "sharp": 0.19670425727963448} +{"time": "09:59:09", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3462.2521352830536, "loss": 0.544837512075901, "entropy": 0.536725714802742, "sharp": 0.25014085322618484} +{"time": "10:00:07", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3463.751641812739, "loss": 0.6907220594584942, "entropy": 0.6240992248058319, "sharp": 0.15954471938312054} +{"time": "10:01:03", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3465.741142535114, "loss": 0.6622186712920666, "entropy": 0.6042149439454079, "sharp": 0.17675238475203514} +{"time": "10:01:58", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3467.4208634795914, "loss": 0.5719126388430595, "entropy": 0.6222634464502335, "sharp": 0.16022830083966255} +{"time": "10:02:18", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.5934485197067261, "noul_T": 0.751309182923458, "noul_auc": 0.7371858702806704, "noul_acc": 0.7, "noul_f1": 0.7807017543859649, "noul_nll": 0.5929632211923599, "noul_ece": 0.09085199332237245, "noul_brier": 0.2027083217451846, "noul_sharp": 0.1448289453983307, "entropy": 0.626939058303833} +{"time": "10:03:15", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3362.314676446188, "loss": 0.6979542076587677, "entropy": 0.621899351477623, "sharp": 0.1505321590229869} +{"time": "10:04:12", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3373.3667478096145, "loss": 0.5715998038649559, "entropy": 0.6176799908280373, "sharp": 0.1588307125493884} +{"time": "10:05:08", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3383.398943703545, "loss": 0.5998402014374733, "entropy": 0.619618771597743, "sharp": 0.15897763427346945} +{"time": "10:06:06", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3390.588459928053, "loss": 0.6634543985128403, "entropy": 0.6531580649316311, "sharp": 0.11612700577825308} +{"time": "10:07:01", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3395.0580963256166, "loss": 0.7021231539547443, "entropy": 0.5879672132432461, "sharp": 0.2023288682103157} +{"time": "10:07:59", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3401.4133537164275, "loss": 0.7054062355309725, "entropy": 0.6439003795385361, "sharp": 0.1311903726309538} +{"time": "10:08:54", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3407.476589087396, "loss": 0.6011508405208588, "entropy": 0.5608271434903145, "sharp": 0.21401756815612316} +{"time": "10:09:50", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3412.5636166316276, "loss": 0.5241282545030117, "entropy": 0.6244533564895391, "sharp": 0.164973015896976} +{"time": "10:10:44", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3415.324107953092, "loss": 0.6702238023281097, "entropy": 0.6285957358777523, "sharp": 0.14554663933813572} +{"time": "10:11:41", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3422.0361041445262, "loss": 0.6391105521470308, "entropy": 0.5985381379723549, "sharp": 0.18845386616885662} +{"time": "10:12:01", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5924770832061768, "noul_T": 0.9455125001007347, "noul_auc": 0.7367301817690841, "noul_acc": 0.682, "noul_f1": 0.7818930041152263, "noul_nll": 0.5949084008336067, "noul_ece": 0.08430295646190647, "noul_brier": 0.20409507125663515, "noul_sharp": 0.17439459264278412, "entropy": 0.6009073257446289} +{"time": "10:12:59", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3367.3975243015957, "loss": 0.7083751261234283, "entropy": 0.6363732665777206, "sharp": 0.13195081800222397} +{"time": "10:13:56", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3373.7500797170105, "loss": 0.6141608692705631, "entropy": 0.621083453297615, "sharp": 0.15805573109537363} +{"time": "10:14:54", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3380.691220485611, "loss": 0.6145672798156738, "entropy": 0.6117670759558678, "sharp": 0.16491932701319456} +{"time": "10:15:50", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3385.384713403154, "loss": 0.6085287630558014, "entropy": 0.6062376871705055, "sharp": 0.1778279859572649} +{"time": "10:16:47", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3389.612210209649, "loss": 0.6004435382783413, "entropy": 0.6042867209762335, "sharp": 0.18555292021483183} +{"time": "10:17:42", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3394.1213418538177, "loss": 0.6062204390764236, "entropy": 0.5940766334533691, "sharp": 0.1961100921034813} +{"time": "10:18:37", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3396.880854672106, "loss": 0.6014404650777578, "entropy": 0.6325034499168396, "sharp": 0.14362389221787453} +{"time": "10:19:34", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3399.100074073732, "loss": 0.586334278807044, "entropy": 0.5943071991205215, "sharp": 0.18739262595772743} +{"time": "10:20:33", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3402.399175903902, "loss": 0.6226176582276821, "entropy": 0.6186786964535713, "sharp": 0.15097895450890064} +{"time": "10:21:29", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3405.467489042052, "loss": 0.6726446505635977, "entropy": 0.5805077478289604, "sharp": 0.19924412108957767} +{"time": "10:21:49", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5799524784088135, "noul_T": 0.883068111002027, "noul_auc": 0.7376246814399757, "noul_acc": 0.712, "noul_f1": 0.7763975155279503, "noul_nll": 0.5832269997298718, "noul_ece": 0.05744416368007661, "noul_brier": 0.19879601933638408, "noul_sharp": 0.1758471131324768, "entropy": 0.598503053188324} +{"time": "10:22:44", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3369.500749975254, "loss": 0.5412260554730892, "entropy": 0.5787854865193367, "sharp": 0.20636413805186749} +{"time": "10:23:42", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3372.728951515247, "loss": 0.6165101639926434, "entropy": 0.6114099845290184, "sharp": 0.15860758814960718} +{"time": "10:24:39", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3375.8728302501336, "loss": 0.6021678745746613, "entropy": 0.5709665045142174, "sharp": 0.21632180735468864} +{"time": "10:25:36", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3379.3745593762524, "loss": 0.6009739991277456, "entropy": 0.6276434976607561, "sharp": 0.14998658746480942} +{"time": "10:26:32", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3381.9702151742413, "loss": 0.62434908375144, "entropy": 0.6268720254302025, "sharp": 0.15082398429512978} +{"time": "10:27:28", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3384.5802794342744, "loss": 0.6606050245463848, "entropy": 0.600510761141777, "sharp": 0.17430782690644264} +{"time": "10:28:25", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3386.559912792454, "loss": 0.5887299813330173, "entropy": 0.5904230875894427, "sharp": 0.19327650591731071} +{"time": "10:29:23", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3390.143172942584, "loss": 0.6204383447766304, "entropy": 0.6256468035280704, "sharp": 0.15091613680124283} +{"time": "10:30:21", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3392.025683794702, "loss": 0.6212443932890892, "entropy": 0.6395474448800087, "sharp": 0.13356971368193626} +{"time": "10:31:15", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3393.4778522972115, "loss": 0.6074783802032471, "entropy": 0.5988033562898636, "sharp": 0.18876321986317635} +{"time": "10:31:35", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5827665328979492, "noul_T": 0.8777426776122027, "noul_auc": 0.738671077281396, "noul_acc": 0.706, "noul_f1": 0.7835051546391752, "noul_nll": 0.5845510936826468, "noul_ece": 0.05825993323326112, "noul_brier": 0.19914310160796378, "noul_sharp": 0.17451563477516174, "entropy": 0.602178692817688} +{"time": "10:32:31", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3367.0241843154345, "loss": 0.6223461180925369, "entropy": 0.5931613519787788, "sharp": 0.1982572916895151} +{"time": "10:33:27", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3369.0735488798096, "loss": 0.6493937186896801, "entropy": 0.6122419163584709, "sharp": 0.1712731560692191} +{"time": "10:34:24", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3371.2727783793375, "loss": 0.6271458640694618, "entropy": 0.6020262017846107, "sharp": 0.1861561331897974} +{"time": "10:35:21", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3373.082842085422, "loss": 0.6439745966345072, "entropy": 0.6227319166064262, "sharp": 0.15354125387966633} +{"time": "10:36:17", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3374.448988763809, "loss": 0.6012121811509132, "entropy": 0.6128716766834259, "sharp": 0.16796387173235416} +{"time": "10:37:14", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3376.1601946920787, "loss": 0.604939267039299, "entropy": 0.6169973909854889, "sharp": 0.16437110304832458} +{"time": "10:38:09", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3378.88191329674, "loss": 0.7051606886088848, "entropy": 0.6473040655255318, "sharp": 0.11803706735372543} +{"time": "10:39:07", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3380.1729161031712, "loss": 0.6193881034851074, "entropy": 0.6173325553536415, "sharp": 0.16723563242703676} +{"time": "10:40:02", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3383.210554585449, "loss": 0.5798128768801689, "entropy": 0.6280985325574875, "sharp": 0.14609775505959988} +{"time": "10:40:58", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3384.838173562786, "loss": 0.570399266667664, "entropy": 0.6086680814623833, "sharp": 0.17789901979267597} +{"time": "10:41:18", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5828287601470947, "noul_T": 0.874907998326825, "noul_auc": 0.7372196249852323, "noul_acc": 0.71, "noul_f1": 0.7864506627393225, "noul_nll": 0.5854806583523751, "noul_ece": 0.054653770446777356, "noul_brier": 0.19942480560856746, "noul_sharp": 0.1742660105228424, "entropy": 0.6026550531387329} +{"time": "10:42:01", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9813814, "tok_per_s": 3313.573213319746, "rows_per_s": 5.686596466348972, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5696535110473633, "noul_T": 0.902481163861105, "choice_ce": 0.6680582165718079, "choice_T": 1.0356540084892951, "score_ce": 1.0546555519104004, "score_T": 0.9725893429803429, "noul_auc": 0.7243605333333334, "noul_acc": 0.703, "noul_f1": 0.7889125799573561, "noul_nll": 0.5870741973295808, "noul_ece": 0.04549183717370033, "noul_brier": 0.20079337328284533, "noul_sharp": 0.18132315576076508, "entropy": 0.5971601605415344}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-E30-b-rw-syn/DONE b/e28/S1-E28-E30-b-rw-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..7328754d3f58694002a472f7afebd62b00e4166b --- /dev/null +++ b/e28/S1-E28-E30-b-rw-syn/DONE @@ -0,0 +1 @@ +2026-10-21 09:26:15 \ No newline at end of file diff --git a/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..d9f140fd7182033a070186cab4720e82f6e983b3 --- /dev/null +++ b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "out_proj", + "k_proj", + "in_proj_z", + "v_proj", + "up_proj", + "gate_proj", + "down_proj", + "in_proj_qkv", + "q_proj", + "o_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7c903f103f5c3d7eab07f5ed38b5e81844f0e98a --- /dev/null +++ b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0ee50bebea0b88edd7c5f872c072f6c13f17f4b23863caf8d7fb0889a055c4ef +size 160486456 diff --git a/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b58d93bc94fc1c4177b0808276fd7c759cc81cc9 --- /dev/null +++ b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11987139cd7bdf9d32a7004966f033a08d93a5836c57229d143dee1daf88f15d +size 176641832 diff --git a/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..ff98c2d12f5f4629b38e7e736139fb2b5f57eb4d --- /dev/null +++ b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.00035553850466385484, + -0.00035579552059061825, + -0.002260262379422784, + -0.0015018812846392393, + -0.00048384504043497145, + 0.00207708403468132, + 0.0016492396825924516, + -0.0021215537562966347, + -0.000633381656371057, + -0.0006318641826510429, + 3.8325008063111454e-05, + 0.00048488526954315603, + 0.00021413616195786744, + -0.0035404094960540533, + -0.004269149620085955, + -0.00523768737912178, + -0.005989670287817717, + -0.004566085524857044, + -0.004969808738678694, + -0.005063395481556654, + -0.0037965441588312387, + -0.0034527042880654335, + -0.005784549284726381, + -0.0035476109478622675 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-rw-syn/ckpt-16000/calibration.json b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..7250d60f760cbf24591e74d04a6b3a085a761411 --- /dev/null +++ b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.8382791588385368, + "choice": 0.9686506632939115, + "score": 1.0137922178515777 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.8382791588385368, + "ce_before": 0.5741623640060425, + "ce_after": 0.5718194842338562 + }, + "choice": { + "n": 60, + "T": 0.9686506632939115, + "ce_before": 0.6623018980026245, + "ce_after": 0.6620704531669617 + }, + "score": { + "n": 53, + "T": 1.0137922178515777, + "ce_before": 1.054081916809082, + "ce_after": 1.0540289878845215 + } + }, + "source": "models/e28/S1-E28-E30-b-rw-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-rw-syn/ckpt-16000/head.safetensors b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..46b58386acae79b570b12db72d37f5ef6f3df3d3 --- /dev/null +++ b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c64a6ccc73f0744a9454809d2d9e3748ce7ae043d6c2d79a61e1511907f53ac +size 393464 diff --git a/e28/S1-E28-E30-b-rw-syn/ckpt-16000/judge_config.json b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..d66190065bbf57a7c9af527a8f8150f160734dfe --- /dev/null +++ b/e28/S1-E28-E30-b-rw-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-E30-b-rw-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-E30-b-rw", + "data_manifest": { + "arm": "E28-E30-b-rw", + "name": null, + "train_sha256": "cb30294ba98bb869fdcaacebbf43f90789f92bfae81c1e60178330f3b8b1df4f", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9900160, + "tok_per_s": 3319.9097091689296, + "rows_per_s": 5.6477793577134205, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5741623640060425, + "noul_T": 0.8382791588385368, + "choice_ce": 0.6623018980026245, + "choice_T": 0.9686506632939115, + "score_ce": 1.054081916809082, + "score_T": 1.0137922178515777, + "noul_auc": 0.7156736, + "noul_acc": 0.686, + "noul_f1": 0.7653213751868461, + "noul_nll": 0.5926208108663559, + "noul_ece": 0.044586173862218866, + "noul_brier": 0.20328538649815972, + "noul_sharp": 0.16902607679367065, + "entropy": 0.6107732057571411 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-rw-syn/train_log.jsonl b/e28/S1-E28-E30-b-rw-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..5cf929dc6397d70e950432f5bfac90c77de9c1a5 --- /dev/null +++ b/e28/S1-E28-E30-b-rw-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "08:36:32", "event": "start", "data": "data/e28/E28-E30-b-rw", "data_manifest": {"arm": "E28-E30-b-rw", "name": null, "train_sha256": "cb30294ba98bb869fdcaacebbf43f90789f92bfae81c1e60178330f3b8b1df4f", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.307108402252197} +{"time": "08:37:29", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3401.522345569772, "loss": 0.5883438028395176, "entropy": 0.649910070002079, "sharp": 0.13027036003768444} +{"time": "08:38:25", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3443.0729229963094, "loss": 0.6410267390310764, "entropy": 0.5527986735105515, "sharp": 0.23993315547704697} +{"time": "08:39:21", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3450.248543776323, "loss": 0.6694338582456112, "entropy": 0.6697213277220726, "sharp": 0.08455516863614321} +{"time": "08:40:17", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3455.6536208050625, "loss": 0.6157706584781408, "entropy": 0.6396417319774628, "sharp": 0.13161780964583158} +{"time": "08:41:13", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3468.720468877729, "loss": 0.6596298180520535, "entropy": 0.5955438911914825, "sharp": 0.1959834024310112} +{"time": "08:42:09", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3471.152767967397, "loss": 0.5875795651227236, "entropy": 0.605655275285244, "sharp": 0.170285876840353} +{"time": "08:43:08", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3476.505253530525, "loss": 0.6192992273718119, "entropy": 0.4956628642976284, "sharp": 0.26945515163242817} +{"time": "08:44:04", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3475.469541408692, "loss": 0.6228486206382513, "entropy": 0.6164748668670654, "sharp": 0.18821429274976254} +{"time": "08:45:02", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3472.7994261571976, "loss": 0.6552108824253082, "entropy": 0.6847280040383339, "sharp": 0.053533174097537994} +{"time": "08:45:58", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3471.0280478612012, "loss": 0.6818425841629505, "entropy": 0.5334899872541428, "sharp": 0.25088020600378513} +{"time": "08:46:18", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.5991434454917908, "noul_T": 1.288247520419859, "noul_auc": 0.7365951629508364, "noul_acc": 0.68, "noul_f1": 0.7796143250688705, "noul_nll": 0.5986181441396475, "noul_ece": 0.09075815457105636, "noul_brier": 0.2053876545875678, "noul_sharp": 0.23110778629779816, "entropy": 0.5446556210517883} +{"time": "08:47:15", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3361.976118425592, "loss": 0.6362625509500504, "entropy": 0.6569162607192993, "sharp": 0.09868691582232714} +{"time": "08:48:10", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3370.4318864500765, "loss": 0.5814001858234406, "entropy": 0.6297832876443863, "sharp": 0.1426638327538967} +{"time": "08:49:06", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3380.4923645252393, "loss": 0.5769799798727036, "entropy": 0.6017874479293823, "sharp": 0.18359841872006655} +{"time": "08:50:03", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3389.4347993362026, "loss": 0.6450922675430775, "entropy": 0.5867063328623772, "sharp": 0.2104090303182602} +{"time": "08:51:01", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3394.2329978834987, "loss": 0.6571156196296215, "entropy": 0.6562857925891876, "sharp": 0.10908288415521383} +{"time": "08:51:59", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3401.534301833461, "loss": 0.6711053252220154, "entropy": 0.6450290270149708, "sharp": 0.12942496594041586} +{"time": "08:52:54", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3403.318669128783, "loss": 0.6606200858950615, "entropy": 0.6219481080770493, "sharp": 0.14433051366358995} +{"time": "08:53:51", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3407.812666986214, "loss": 0.638141542673111, "entropy": 0.6622077487409115, "sharp": 0.10792828351259232} +{"time": "08:54:48", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3411.3401723861775, "loss": 0.615454588085413, "entropy": 0.6276964172720909, "sharp": 0.1525723710656166} +{"time": "08:55:47", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3415.4407044669415, "loss": 0.6039196774363518, "entropy": 0.5994380861520767, "sharp": 0.1847896594554186} +{"time": "08:56:06", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5863097906112671, "noul_T": 0.8880723412125955, "noul_auc": 0.7325614757556833, "noul_acc": 0.702, "noul_f1": 0.7711213517665131, "noul_nll": 0.5859597466289997, "noul_ece": 0.06177112895250322, "noul_brier": 0.19995281539220022, "noul_sharp": 0.17855989933013916, "entropy": 0.604243278503418} +{"time": "08:57:03", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3363.886893231434, "loss": 0.6876555532217026, "entropy": 0.5639746524393559, "sharp": 0.23872085474431515} +{"time": "08:58:01", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3368.414484263588, "loss": 0.6542440615594387, "entropy": 0.5928572341799736, "sharp": 0.19763751327991486} +{"time": "08:58:59", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3373.493083296255, "loss": 0.6252200752496719, "entropy": 0.5814718082547188, "sharp": 0.2116015050560236} +{"time": "08:59:58", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3379.63794309933, "loss": 0.5984925776720047, "entropy": 0.6479922775179148, "sharp": 0.1318037435412407} +{"time": "09:00:53", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3384.6915920635, "loss": 0.6469800099730492, "entropy": 0.5924363061785698, "sharp": 0.18962757103145123} +{"time": "09:01:51", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3388.663559184963, "loss": 0.6245180666446686, "entropy": 0.6358210667967796, "sharp": 0.1537108551710844} +{"time": "09:02:48", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3391.5071859364743, "loss": 0.6238603834062815, "entropy": 0.6437391974031925, "sharp": 0.13688329234719276} +{"time": "09:03:43", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3395.7515197492176, "loss": 0.6042768806219101, "entropy": 0.5957635417580605, "sharp": 0.1871871231123805} +{"time": "09:04:40", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3398.4534141950708, "loss": 0.6525414288043976, "entropy": 0.6323323305696249, "sharp": 0.14127254765480757} +{"time": "09:05:37", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3402.0071265625147, "loss": 0.6971588358283043, "entropy": 0.5999293550848961, "sharp": 0.19242287427186966} +{"time": "09:05:57", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5822479128837585, "noul_T": 0.9021554293144558, "noul_auc": 0.7371521155761084, "noul_acc": 0.682, "noul_f1": 0.7527216174183515, "noul_nll": 0.5834956202208996, "noul_ece": 0.08405899977684021, "noul_brier": 0.19886568784441258, "noul_sharp": 0.18089844286441803, "entropy": 0.5982798337936401} +{"time": "09:06:55", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3367.246937193753, "loss": 0.6258633881807327, "entropy": 0.6440308317542076, "sharp": 0.13705564104020596} +{"time": "09:07:52", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3371.6449059428687, "loss": 0.6225789897143841, "entropy": 0.6264785341918468, "sharp": 0.15325424820184708} +{"time": "09:08:49", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3375.0459306764615, "loss": 0.6210546270012856, "entropy": 0.6218028105795383, "sharp": 0.1625778879970312} +{"time": "09:09:46", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3378.4908749727488, "loss": 0.658441074192524, "entropy": 0.6416617557406425, "sharp": 0.13657623808830976} +{"time": "09:10:41", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3380.8601886158067, "loss": 0.6781150251626968, "entropy": 0.6120394244790077, "sharp": 0.16070055682212114} +{"time": "09:11:38", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3385.23288013647, "loss": 0.630499541759491, "entropy": 0.6040371209383011, "sharp": 0.17993340454995632} +{"time": "09:12:32", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3388.1383791439353, "loss": 0.6200841888785362, "entropy": 0.5992672070860863, "sharp": 0.18843172676861286} +{"time": "09:13:29", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3390.472869983647, "loss": 0.6384056778624654, "entropy": 0.6245409157127142, "sharp": 0.15997616946697235} +{"time": "09:14:25", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3392.8680246934246, "loss": 0.615900531411171, "entropy": 0.6180580481886864, "sharp": 0.17485034465789795} +{"time": "09:15:23", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3395.2875262837892, "loss": 0.6269356422126293, "entropy": 0.6206650622189045, "sharp": 0.16762340813875198} +{"time": "09:15:42", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5870345234870911, "noul_T": 0.743699934200717, "noul_auc": 0.7336753810062276, "noul_acc": 0.7, "noul_f1": 0.7740963855421686, "noul_nll": 0.5895199077129364, "noul_ece": 0.06320365321636201, "noul_brier": 0.201227247726989, "noul_sharp": 0.15314294397830963, "entropy": 0.6236692070960999} +{"time": "09:16:40", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3370.2978863474486, "loss": 0.6348151676356792, "entropy": 0.6186058912426233, "sharp": 0.15421402174979448} +{"time": "09:17:35", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3373.637803214827, "loss": 0.6177404597401619, "entropy": 0.6045624725520611, "sharp": 0.18852545134723186} +{"time": "09:18:32", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3376.538766168761, "loss": 0.6765347924083471, "entropy": 0.6142625203356147, "sharp": 0.16706210002303123} +{"time": "09:19:30", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3378.4574246041093, "loss": 0.6313289701938629, "entropy": 0.611705219373107, "sharp": 0.16677995584905148} +{"time": "09:20:27", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3381.6819019760615, "loss": 0.593661442399025, "entropy": 0.6065301150083542, "sharp": 0.1769587341696024} +{"time": "09:21:23", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3383.3434019041897, "loss": 0.6441304190084338, "entropy": 0.629310604184866, "sharp": 0.1496828319504857} +{"time": "09:22:20", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3384.9121822885195, "loss": 0.6589799635112286, "entropy": 0.6449206918478012, "sharp": 0.12577264942228794} +{"time": "09:23:17", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3386.965127950355, "loss": 0.5903299860656261, "entropy": 0.623900406062603, "sharp": 0.156180951744318} +{"time": "09:24:15", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3389.2051791471026, "loss": 0.6339425444602966, "entropy": 0.6186253726482391, "sharp": 0.1701121050864458} +{"time": "09:25:12", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3390.799680825442, "loss": 0.5940200723707676, "entropy": 0.6235075742006302, "sharp": 0.163092702627182} +{"time": "09:25:32", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5874089598655701, "noul_T": 0.8182190910381258, "noul_auc": 0.7323589475283117, "noul_acc": 0.694, "noul_f1": 0.7627906976744186, "noul_nll": 0.5888926620781422, "noul_ece": 0.050881282746791844, "noul_brier": 0.20107069533652622, "noul_sharp": 0.16376787424087524, "entropy": 0.6145408749580383} +{"time": "09:26:15", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9900160, "tok_per_s": 3319.9097091689296, "rows_per_s": 5.6477793577134205, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5741623640060425, "noul_T": 0.8382791588385368, "choice_ce": 0.6623018980026245, "choice_T": 0.9686506632939115, "score_ce": 1.054081916809082, "score_T": 1.0137922178515777, "noul_auc": 0.7156736, "noul_acc": 0.686, "noul_f1": 0.7653213751868461, "noul_nll": 0.5926208108663559, "noul_ece": 0.044586173862218866, "noul_brier": 0.20328538649815972, "noul_sharp": 0.16902607679367065, "entropy": 0.6107732057571411}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-E30-b-uni-syn/DONE b/e28/S1-E28-E30-b-uni-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..377386fa0a88ab1306fa7bdbe38e8a9b5346d17e --- /dev/null +++ b/e28/S1-E28-E30-b-uni-syn/DONE @@ -0,0 +1 @@ +2026-10-21 08:10:20 \ No newline at end of file diff --git a/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..c7cd229d6e93083c92481cc76a16c9e711868ef3 --- /dev/null +++ b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "in_proj_qkv", + "up_proj", + "o_proj", + "v_proj", + "gate_proj", + "q_proj", + "out_proj", + "k_proj", + "in_proj_z", + "down_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8ccb6e8067956024246af0b902033d95edc1c4e5 --- /dev/null +++ b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:62e63a21220dad3ad34785d10ef0e3f8dcd498b14e912a4942607f379108835d +size 160486456 diff --git a/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..cff4611de50fc67970319820204023c4576c13f1 --- /dev/null +++ b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ce95a8ab08dab4b978f8daa5a64c5765777f3b6a0489337d7d8eba8ac67bb1dc +size 176641832 diff --git a/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..70a86f9f4a6e14c271be914161b6d08870724870 --- /dev/null +++ b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.00037657454959116876, + -0.0003768277820199728, + -0.0020475457422435284, + -0.0013714126544073224, + -0.0002748112310655415, + 0.0023842831142246723, + 0.0017594359815120697, + -0.002070085611194372, + -0.0007267724722623825, + -0.0006606851820833981, + -0.0002832302125170827, + 0.00010186276631429791, + 0.0002777380868792534, + -0.003158177714794874, + -0.004123254679143429, + -0.0050786929205060005, + -0.0059461393393576145, + -0.004339568316936493, + -0.004928252659738064, + -0.004940286744385958, + -0.00389523571357131, + -0.003354681422933936, + -0.005609809421002865, + -0.0034640179947018623 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-uni-syn/ckpt-16000/calibration.json b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..f7fc8609ec6213b7621066f92b53729f3fb54c49 --- /dev/null +++ b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.7408984579354347, + "choice": 1.0667111239077942, + "score": 0.9784231432202636 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.7408984579354347, + "ce_before": 0.5773091912269592, + "ce_after": 0.5711639523506165 + }, + "choice": { + "n": 60, + "T": 1.0667111239077942, + "ce_before": 0.6674884557723999, + "ce_after": 0.666592001914978 + }, + "score": { + "n": 53, + "T": 0.9784231432202636, + "ce_before": 1.0595662593841553, + "ce_after": 1.059427261352539 + } + }, + "source": "models/e28/S1-E28-E30-b-uni-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-uni-syn/ckpt-16000/head.safetensors b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..654f203788eba982fdbfa5e8ddab4c88672a33f2 --- /dev/null +++ b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9da37e3d08f1d5ee32996c1eacf3abc0c29f85fb2509fd8563f3e9c6da38e6b5 +size 393464 diff --git a/e28/S1-E28-E30-b-uni-syn/ckpt-16000/judge_config.json b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..fc5bdcd2fa680139c8c940ebe8079add7435a300 --- /dev/null +++ b/e28/S1-E28-E30-b-uni-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-E30-b-uni-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-E30-b-uni", + "data_manifest": { + "arm": "E28-E30-b-uni", + "name": null, + "train_sha256": "f5d01f0d68c894885b7a90f9f22bd228eb0892043e113864a5ec13d221dfcb74", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9882729, + "tok_per_s": 3328.6569209480194, + "rows_per_s": 5.6726476893772375, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5773091912269592, + "noul_T": 0.7408984579354347, + "choice_ce": 0.6674884557723999, + "choice_T": 1.0667111239077942, + "score_ce": 1.0595662593841553, + "score_T": 0.9784231432202636, + "noul_auc": 0.7157504, + "noul_acc": 0.69, + "noul_f1": 0.7658610271903323, + "noul_nll": 0.5961172372102738, + "noul_ece": 0.06159302225708963, + "noul_brier": 0.20454293562295356, + "noul_sharp": 0.15185491740703583, + "entropy": 0.626173734664917 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-E30-b-uni-syn/train_log.jsonl b/e28/S1-E28-E30-b-uni-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..07e8b7014f83f20e3eb37cfb0d5e3191a45ef49f --- /dev/null +++ b/e28/S1-E28-E30-b-uni-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "07:20:50", "event": "start", "data": "data/e28/E28-E30-b-uni", "data_manifest": {"arm": "E28-E30-b-uni", "name": null, "train_sha256": "f5d01f0d68c894885b7a90f9f22bd228eb0892043e113864a5ec13d221dfcb74", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.58167028427124} +{"time": "07:21:48", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3412.84453176655, "loss": 0.6643712073564529, "entropy": 0.6881146878004074, "sharp": 0.04014213290065527} +{"time": "07:22:46", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3454.5847470291737, "loss": 0.6722264401614666, "entropy": 0.6747759338468313, "sharp": 0.07759428303688765} +{"time": "07:23:41", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3464.080181981956, "loss": 0.7008826434612274, "entropy": 0.680109292268753, "sharp": 0.06770142633467913} +{"time": "07:24:39", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3472.614519716595, "loss": 0.6368321049958467, "entropy": 0.6733085364103317, "sharp": 0.07902656309306622} +{"time": "07:25:35", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3484.412773934003, "loss": 0.6529328739270568, "entropy": 0.6374943032860756, "sharp": 0.14144005347043276} +{"time": "07:26:32", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3490.7625975206442, "loss": 0.6597438417375088, "entropy": 0.632667750120163, "sharp": 0.14504528231918812} +{"time": "07:27:27", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3493.0948752724894, "loss": 0.6349328346550465, "entropy": 0.6384568735957146, "sharp": 0.13547073397785425} +{"time": "07:28:24", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3491.403507837829, "loss": 0.6893104016780853, "entropy": 0.6736803203821182, "sharp": 0.07898853998631239} +{"time": "07:29:19", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3491.648073128078, "loss": 0.6984788179397583, "entropy": 0.6669861674308777, "sharp": 0.08741604909300804} +{"time": "07:30:16", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3488.417808169901, "loss": 0.67269217222929, "entropy": 0.6663933768868446, "sharp": 0.08974896930158138} +{"time": "07:30:35", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.5960289835929871, "noul_T": 0.7811769540400487, "noul_auc": 0.7279539585829774, "noul_acc": 0.674, "noul_f1": 0.7776261937244202, "noul_nll": 0.6027385316342115, "noul_ece": 0.1022030407190323, "noul_brier": 0.20724945352805735, "noul_sharp": 0.1460534632205963, "entropy": 0.6249595880508423} +{"time": "07:31:32", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3381.315259676401, "loss": 0.6385923810303211, "entropy": 0.6601524725556374, "sharp": 0.09805411286652088} +{"time": "07:32:28", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3389.921180449784, "loss": 0.6750622615218163, "entropy": 0.6181630715727806, "sharp": 0.1618830393999815} +{"time": "07:33:27", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3400.861263903481, "loss": 0.6404179111123085, "entropy": 0.6749487146735191, "sharp": 0.07947001047432423} +{"time": "07:34:24", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3403.014154177521, "loss": 0.637463815510273, "entropy": 0.6444430574774742, "sharp": 0.1324672568589449} +{"time": "07:35:21", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3407.979621887157, "loss": 0.610202451236546, "entropy": 0.6256229765713215, "sharp": 0.1523785488680005} +{"time": "07:36:17", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3413.4175502783255, "loss": 0.7155324220657349, "entropy": 0.6795186437666416, "sharp": 0.06899222917854786} +{"time": "07:37:14", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3418.7640759889823, "loss": 0.6590254902839661, "entropy": 0.659384373575449, "sharp": 0.11563058942556381} +{"time": "07:38:11", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3424.189911328455, "loss": 0.593407578766346, "entropy": 0.6386690959334373, "sharp": 0.135503220371902} +{"time": "07:39:08", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3428.606238631469, "loss": 0.6568507980555296, "entropy": 0.6741286143660545, "sharp": 0.07894595246762037} +{"time": "07:40:06", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3432.909847583256, "loss": 0.6499505769461393, "entropy": 0.6766723170876503, "sharp": 0.07397074531763792} +{"time": "07:40:26", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.6239725947380066, "noul_T": 0.6132484938268201, "noul_auc": 0.728274628276316, "noul_acc": 0.692, "noul_f1": 0.7372013651877133, "noul_nll": 0.6211362561285496, "noul_ece": 0.10925235217809678, "noul_brier": 0.2151243131076792, "noul_sharp": 0.10683318972587585, "entropy": 0.6592355370521545} +{"time": "07:41:22", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3378.807620455537, "loss": 0.7824937254190445, "entropy": 0.6668129041790962, "sharp": 0.09836865495890379} +{"time": "07:42:20", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3384.4931737289567, "loss": 0.6671350710093975, "entropy": 0.6606801003217697, "sharp": 0.10917341336607933} +{"time": "07:43:16", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3388.472758125226, "loss": 0.6623816471546888, "entropy": 0.6674660854041576, "sharp": 0.08345185220241547} +{"time": "07:44:13", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3394.409281106294, "loss": 0.6756945550441742, "entropy": 0.6593797355890274, "sharp": 0.10428703296929598} +{"time": "07:45:09", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3397.737375610196, "loss": 0.6887680180370808, "entropy": 0.6785908713936806, "sharp": 0.06562523916363716} +{"time": "07:46:04", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3400.2634676963544, "loss": 0.6339290328323841, "entropy": 0.6490484178066254, "sharp": 0.12375408597290516} +{"time": "07:47:01", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3403.1155760418164, "loss": 0.6234707906842232, "entropy": 0.6582430452108383, "sharp": 0.10739425756037235} +{"time": "07:47:58", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3408.3727934182716, "loss": 0.6599689871072769, "entropy": 0.6678161956369877, "sharp": 0.09251843299716711} +{"time": "07:48:53", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3411.6270132321597, "loss": 0.6376436054706573, "entropy": 0.6270197257399559, "sharp": 0.15573505219072104} +{"time": "07:49:48", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3414.7073858374806, "loss": 0.6898017637431622, "entropy": 0.6396807972341776, "sharp": 0.14090069942176342} +{"time": "07:50:08", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.592384934425354, "noul_T": 0.7832765252986107, "noul_auc": 0.7269750721506809, "noul_acc": 0.71, "noul_f1": 0.779969650986343, "noul_nll": 0.592866287112236, "noul_ece": 0.06744354343414308, "noul_brier": 0.20230756666763247, "noul_sharp": 0.15819570422172546, "entropy": 0.6233943104743958} +{"time": "07:51:04", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3377.963409952961, "loss": 0.6560590118169785, "entropy": 0.6759481653571129, "sharp": 0.07566426042467356} +{"time": "07:52:00", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3381.8816390302327, "loss": 0.6521664895117283, "entropy": 0.6562619507312775, "sharp": 0.11154479160904884} +{"time": "07:52:55", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3385.7745612506787, "loss": 0.6142350137233734, "entropy": 0.6497357785701752, "sharp": 0.12369169667363167} +{"time": "07:53:52", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3389.070081935378, "loss": 0.721906416118145, "entropy": 0.6396926231682301, "sharp": 0.12998747546225786} +{"time": "07:54:50", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3390.81551865858, "loss": 0.6273990049958229, "entropy": 0.6704195290803909, "sharp": 0.08674909919500351} +{"time": "07:55:46", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3393.829900525542, "loss": 0.6839748788625002, "entropy": 0.6553663816303015, "sharp": 0.1161989001557231} +{"time": "07:56:43", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3398.4252797052277, "loss": 0.6655334010720253, "entropy": 0.6360623091459274, "sharp": 0.1377185545861721} +{"time": "07:57:41", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3401.5580002975503, "loss": 0.6637069880962372, "entropy": 0.6584352217614651, "sharp": 0.11270937602967024} +{"time": "07:58:36", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3404.268115644706, "loss": 0.6711983382701874, "entropy": 0.6425489857792854, "sharp": 0.1311301663517952} +{"time": "07:59:34", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3407.533151282695, "loss": 0.6224565468728542, "entropy": 0.6446594595909119, "sharp": 0.132312529720366} +{"time": "07:59:54", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5929962992668152, "noul_T": 0.7253094590302291, "noul_auc": 0.730164891731785, "noul_acc": 0.708, "noul_f1": 0.7667731629392971, "noul_nll": 0.5956401140987873, "noul_ece": 0.06600068140029908, "noul_brier": 0.20373952778124688, "noul_sharp": 0.14718371629714966, "entropy": 0.630206823348999} +{"time": "08:00:51", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3380.775547383758, "loss": 0.66121406853199, "entropy": 0.6640224754810333, "sharp": 0.09943372011184692} +{"time": "08:01:49", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3383.702964826947, "loss": 0.6427838280797005, "entropy": 0.6557530388236046, "sharp": 0.11084597278386354} +{"time": "08:02:43", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3385.318776343885, "loss": 0.7007791735231876, "entropy": 0.6650380529463291, "sharp": 0.09824907127767801} +{"time": "08:03:41", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3387.226194680698, "loss": 0.710824653506279, "entropy": 0.6376304924488068, "sharp": 0.1433349335566163} +{"time": "08:04:37", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3389.1983481771895, "loss": 0.6227287873625755, "entropy": 0.643506683409214, "sharp": 0.1370569234713912} +{"time": "08:05:33", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3391.4489286801927, "loss": 0.6685913428664207, "entropy": 0.6697678864002228, "sharp": 0.08324440754950047} +{"time": "08:06:33", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3393.5461792859783, "loss": 0.6908168271183968, "entropy": 0.6579391732811928, "sharp": 0.1065461216494441} +{"time": "08:07:28", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3395.6029137369023, "loss": 0.6634692773222923, "entropy": 0.6619432717561722, "sharp": 0.09973124507814646} +{"time": "08:08:22", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3398.1314947693018, "loss": 0.6402052715420723, "entropy": 0.6511539071798325, "sharp": 0.11870834976434708} +{"time": "08:09:17", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3400.056035986152, "loss": 0.6373008638620377, "entropy": 0.6451828852295876, "sharp": 0.13431397639214993} +{"time": "08:09:37", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5930728316307068, "noul_T": 0.7190314816661275, "noul_auc": 0.7314306931528581, "noul_acc": 0.706, "noul_f1": 0.768503937007874, "noul_nll": 0.5944534106254578, "noul_ece": 0.06491368138790131, "noul_brier": 0.20319309121618032, "noul_sharp": 0.14663207530975342, "entropy": 0.6308554410934448} +{"time": "08:10:20", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9882729, "tok_per_s": 3328.6569209480194, "rows_per_s": 5.6726476893772375, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5773091912269592, "noul_T": 0.7408984579354347, "choice_ce": 0.6674884557723999, "choice_T": 1.0667111239077942, "score_ce": 1.0595662593841553, "score_T": 0.9784231432202636, "noul_auc": 0.7157504, "noul_acc": 0.69, "noul_f1": 0.7658610271903323, "noul_nll": 0.5961172372102738, "noul_ece": 0.06159302225708963, "noul_brier": 0.20454293562295356, "noul_sharp": 0.15185491740703583, "entropy": 0.626173734664917}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-E30-e18-rw-syn/DONE b/e28/S1-E28-E30-e18-rw-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..739d347cf2c3f05a3b83aeb3d8c27b4c61a6e8e4 --- /dev/null +++ b/e28/S1-E28-E30-e18-rw-syn/DONE @@ -0,0 +1 @@ +2026-10-21 16:14:24 \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..b85dbece762d670211d4e2c6c4fbdf19745cb6fe --- /dev/null +++ b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "q_proj", + "up_proj", + "k_proj", + "gate_proj", + "in_proj_qkv", + "down_proj", + "in_proj_z", + "out_proj", + "o_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9e2fc6fc316e9cb651e4008f77c56bb56fe43cf5 --- /dev/null +++ b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5e03696b9ad62360dfd30a1e7aa1c3a49836a156b6b11bcb5d0d96ebf1323731 +size 160486456 diff --git a/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8b0517b13f8ace187fb30700b6ff4d8213218f04 --- /dev/null +++ b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9ac488f1076d6f4593b491eb59bf017dd3a26767323b28416a7fedb9f210798c +size 176641832 diff --git a/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..2aa987554efdfeeab207e2cad8320a3dfbd5ef09 --- /dev/null +++ b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.0003756797523237765, + -0.0003759448300115764, + -0.0029805691447108984, + -0.0017459788359701633, + -0.0004286786715965718, + 0.0022440911270678043, + 0.0014619691064581275, + -0.002868198323994875, + -0.0004240291309542954, + -0.0006189714185893536, + -3.2744937925599515e-05, + 0.0004517995985224843, + 0.00044472035369835794, + -0.0040838224813342094, + -0.004394893534481525, + -0.005323839373886585, + -0.005997456144541502, + -0.004507772624492645, + -0.0050461567007005215, + -0.005391790997236967, + -0.004566457588225603, + -0.003841257654130459, + -0.005852279718965292, + -0.003864402649924159 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/calibration.json b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..0bb56be473ceefb482c4a0b9268dc67fd87e384c --- /dev/null +++ b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.8376916311270307, + "choice": 1.0467250894734454, + "score": 0.99517842781877 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.8376916311270307, + "ce_before": 0.5578375458717346, + "ce_after": 0.5552768111228943 + }, + "choice": { + "n": 60, + "T": 1.0467250894734454, + "ce_before": 0.6999159455299377, + "ce_after": 0.699444055557251 + }, + "score": { + "n": 53, + "T": 0.99517842781877, + "ce_before": 1.0730246305465698, + "ce_after": 1.073015809059143 + } + }, + "source": "models/e28/S1-E28-E30-e18-rw-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/head.safetensors b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8c372239e31029fc9167c55b58f8c67f40e277b9 --- /dev/null +++ b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:00acdfd0710a1836269cf7471416344557b0b7a45b9fd3a7da4546c6c72f9c8b +size 393464 diff --git a/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/judge_config.json b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..fe361d482e05d57216408224a91e85409afe0731 --- /dev/null +++ b/e28/S1-E28-E30-e18-rw-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-E30-e18-rw-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-E30-e18-rw", + "data_manifest": { + "arm": "E28-E30-e18-rw", + "name": null, + "train_sha256": "ebc5c0860c98a1cafb3b765e2d0a17e35d4075f0cdea6b0822318ac8cc82155f", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 12630995, + "tok_per_s": 3391.972358669959, + "rows_per_s": 4.522810628283892, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5578375458717346, + "noul_T": 0.8376916311270307, + "choice_ce": 0.6999159455299377, + "choice_T": 1.0467250894734454, + "score_ce": 1.0730246305465698, + "score_T": 0.99517842781877, + "noul_auc": 0.7416917333333333, + "noul_acc": 0.698, + "noul_f1": 0.7772861356932154, + "noul_nll": 0.5728344293050468, + "noul_ece": 0.03640232239663601, + "noul_brier": 0.19540384433301067, + "noul_sharp": 0.1810053139925003, + "entropy": 0.5957744717597961 + }, + "dropped_long": 0, + "peak_gib": 18.28459882736206 + } +} \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-rw-syn/train_log.jsonl b/e28/S1-E28-E30-e18-rw-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..08849002eb2ae98e90fc481151869f1645de6876 --- /dev/null +++ b/e28/S1-E28-E30-e18-rw-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "15:12:20", "event": "start", "data": "data/e28/E28-E30-e18-rw", "data_manifest": {"arm": "E28-E30-e18-rw", "name": null, "train_sha256": "ebc5c0860c98a1cafb3b765e2d0a17e35d4075f0cdea6b0822318ac8cc82155f", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.710962295532227} +{"time": "15:13:33", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3498.3590276276814, "loss": 0.6094549149274826, "entropy": 0.6432857289910316, "sharp": 0.13042087107896805} +{"time": "15:14:44", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3520.2299620357335, "loss": 0.6024592909961939, "entropy": 0.608183030039072, "sharp": 0.15863023791462183} +{"time": "15:15:57", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3519.901750256618, "loss": 0.5865871794521809, "entropy": 0.5597596541047096, "sharp": 0.22048311587423086} +{"time": "15:17:08", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3519.9434095082725, "loss": 0.5865075550973415, "entropy": 0.5426506819203496, "sharp": 0.2257890459150076} +{"time": "15:18:21", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3527.5382629808696, "loss": 0.6055621178820729, "entropy": 0.5479004895314574, "sharp": 0.2313134167343378} +{"time": "15:19:34", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3527.270029095542, "loss": 0.5582474917173386, "entropy": 0.5423509106040001, "sharp": 0.20900608133524656} +{"time": "15:20:44", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3529.2654877496234, "loss": 0.5646139122545719, "entropy": 0.6189629398286343, "sharp": 0.1647495273500681} +{"time": "15:21:55", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3529.292958095692, "loss": 0.5865089036524296, "entropy": 0.5801612809300423, "sharp": 0.20726720802485943} +{"time": "15:23:08", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3528.0539789963977, "loss": 0.5808851309120655, "entropy": 0.5424130819737911, "sharp": 0.23135964386165142} +{"time": "15:24:19", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3525.2784591960244, "loss": 0.5701629538089037, "entropy": 0.5811720080673695, "sharp": 0.20955547876656055} +{"time": "15:24:39", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.5752249360084534, "noul_T": 0.8205703188210224, "noul_auc": 0.7481055172064606, "noul_acc": 0.708, "noul_f1": 0.7767584097859327, "noul_nll": 0.5768077117055654, "noul_ece": 0.07252209389209748, "noul_brier": 0.1969883324671751, "noul_sharp": 0.16683779656887054, "entropy": 0.6021084785461426} +{"time": "15:25:51", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3438.964364422293, "loss": 0.558182317763567, "entropy": 0.6008301228284836, "sharp": 0.17831526510417461} +{"time": "15:27:04", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3443.4595827471658, "loss": 0.5407590232789516, "entropy": 0.5331138931214809, "sharp": 0.2461653482168913} +{"time": "15:28:16", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3451.0908923382594, "loss": 0.5531845353543758, "entropy": 0.5524345859885216, "sharp": 0.2192077934741974} +{"time": "15:29:28", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3457.1245867864204, "loss": 0.5915894061326981, "entropy": 0.6048745177686214, "sharp": 0.17468809615820646} +{"time": "15:30:36", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3458.5461239813417, "loss": 0.549721971154213, "entropy": 0.5657339170575142, "sharp": 0.2182680731639266} +{"time": "15:31:47", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3461.3498913619737, "loss": 0.6722522024065256, "entropy": 0.6423698775470257, "sharp": 0.11253105662763119} +{"time": "15:32:59", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3466.956843905102, "loss": 0.5691361017525196, "entropy": 0.565075034275651, "sharp": 0.21118027344346046} +{"time": "15:34:12", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3469.881832141786, "loss": 0.5505273416638374, "entropy": 0.5809369385242462, "sharp": 0.19857155345380306} +{"time": "15:35:21", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3472.6249875477192, "loss": 0.5613527493551373, "entropy": 0.568789154291153, "sharp": 0.20604482851922512} +{"time": "15:36:32", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3474.7296223369203, "loss": 0.5860893055796623, "entropy": 0.5793001558631659, "sharp": 0.1973577393218875} +{"time": "15:36:52", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5755560994148254, "noul_T": 0.853604582174063, "noul_auc": 0.7469409798990734, "noul_acc": 0.7, "noul_f1": 0.7800586510263929, "noul_nll": 0.5780825258940459, "noul_ece": 0.0664496066570282, "noul_brier": 0.19757697211343184, "noul_sharp": 0.17267189919948578, "entropy": 0.5985440611839294} +{"time": "15:38:02", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3432.254256570956, "loss": 0.6074366420507431, "entropy": 0.5857502780854702, "sharp": 0.19984089769423008} +{"time": "15:39:13", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3436.698854039651, "loss": 0.5733276642858982, "entropy": 0.5853805392980576, "sharp": 0.19018685445189476} +{"time": "15:40:24", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3441.110005367277, "loss": 0.5747623983770609, "entropy": 0.5365629866719246, "sharp": 0.2373831458389759} +{"time": "15:41:36", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3445.150115686243, "loss": 0.5456774607300758, "entropy": 0.5443911366164684, "sharp": 0.2313238438218832} +{"time": "15:42:49", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3448.660870672728, "loss": 0.5475920587778091, "entropy": 0.5649570897221565, "sharp": 0.2153997402638197} +{"time": "15:43:59", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3451.0530719521694, "loss": 0.5779330311343074, "entropy": 0.5744932927191257, "sharp": 0.20356783457100391} +{"time": "15:45:11", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3453.6421224306337, "loss": 0.522145114839077, "entropy": 0.5594330504536629, "sharp": 0.22555022034794092} +{"time": "15:46:25", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3456.2062055147826, "loss": 0.6006284765899181, "entropy": 0.5314432345330715, "sharp": 0.23938058130443096} +{"time": "15:47:36", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3458.7551323738967, "loss": 0.6078423745930195, "entropy": 0.5936413295567036, "sharp": 0.18614145275205374} +{"time": "15:48:47", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3460.811375235234, "loss": 0.6090671755373478, "entropy": 0.5761802531778812, "sharp": 0.1969085792079568} +{"time": "15:49:07", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5752264261245728, "noul_T": 0.9050779005036437, "noul_auc": 0.7474810551720648, "noul_acc": 0.688, "noul_f1": 0.7483870967741936, "noul_nll": 0.5772214051038027, "noul_ece": 0.0551212261915207, "noul_brier": 0.1972346361412455, "noul_sharp": 0.18443460762500763, "entropy": 0.590779185295105} +{"time": "15:50:19", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3433.2927397132817, "loss": 0.5460403710603714, "entropy": 0.5514236204326153, "sharp": 0.2316950224339962} +{"time": "15:51:30", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3435.1484710948907, "loss": 0.5659809466451406, "entropy": 0.5620591901242733, "sharp": 0.21866302378475666} +{"time": "15:52:43", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3438.409633838128, "loss": 0.5851531066000462, "entropy": 0.5923633314669132, "sharp": 0.17569941002875566} +{"time": "15:53:54", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3441.306422405839, "loss": 0.5682275239378214, "entropy": 0.5072940848767757, "sharp": 0.24459691531956196} +{"time": "15:55:05", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3443.8314098920273, "loss": 0.5622452348470688, "entropy": 0.5912220329046249, "sharp": 0.17804535944014788} +{"time": "15:56:17", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3446.0294425213992, "loss": 0.6220252066850662, "entropy": 0.5683734640479088, "sharp": 0.21785544231534004} +{"time": "15:57:29", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3449.198519893499, "loss": 0.5583401843905449, "entropy": 0.5478718727827072, "sharp": 0.22655513882637024} +{"time": "15:58:40", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3450.4928006127184, "loss": 0.5875684879720211, "entropy": 0.5812307745218277, "sharp": 0.20919272862374783} +{"time": "15:59:51", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3452.7615760642925, "loss": 0.630669642239809, "entropy": 0.6149725690484047, "sharp": 0.15796271432191133} +{"time": "16:01:02", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3455.035522347482, "loss": 0.6213914286345243, "entropy": 0.5905652921646833, "sharp": 0.1854542624205351} +{"time": "16:01:22", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5701988339424133, "noul_T": 0.8441085505050709, "noul_auc": 0.7499282712528058, "noul_acc": 0.698, "noul_f1": 0.7736131934032984, "noul_nll": 0.5722478541433811, "noul_ece": 0.05561212909221651, "noul_brier": 0.19505827178968455, "noul_sharp": 0.1775791347026825, "entropy": 0.5954534411430359} +{"time": "16:02:33", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3434.273015108374, "loss": 0.5980491191148758, "entropy": 0.5867762416601181, "sharp": 0.18846160545945168} +{"time": "16:03:44", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3436.6975468307005, "loss": 0.625732721760869, "entropy": 0.607760714367032, "sharp": 0.16382859274744987} +{"time": "16:04:57", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3438.3984885628406, "loss": 0.5563508179038763, "entropy": 0.5323305148631334, "sharp": 0.2367895022034645} +{"time": "16:06:09", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3440.413553894043, "loss": 0.6071335915476084, "entropy": 0.5641441829502583, "sharp": 0.20671147666871548} +{"time": "16:07:20", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3443.0356201608342, "loss": 0.5626456923782825, "entropy": 0.5874485410749912, "sharp": 0.18019583076238632} +{"time": "16:08:33", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3444.6479503915753, "loss": 0.5509283831343055, "entropy": 0.5794705608859658, "sharp": 0.20034518651664257} +{"time": "16:09:46", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3445.8432599722714, "loss": 0.5792571492493153, "entropy": 0.5494369156658649, "sharp": 0.22274667490273714} +{"time": "16:10:58", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3447.2486101658437, "loss": 0.5814392864704132, "entropy": 0.5536890402436256, "sharp": 0.2207889100536704} +{"time": "16:12:09", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3448.735186175676, "loss": 0.5556817753240466, "entropy": 0.5775649230927229, "sharp": 0.19082816038280725} +{"time": "16:13:22", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3449.800767941746, "loss": 0.5381961949169636, "entropy": 0.566188020631671, "sharp": 0.20552089996635914} +{"time": "16:13:42", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5710093379020691, "noul_T": 0.8307959114665975, "noul_auc": 0.7500464127187727, "noul_acc": 0.692, "noul_f1": 0.764525993883792, "noul_nll": 0.5733713551685214, "noul_ece": 0.04564259189367296, "noul_brier": 0.19530651996758752, "noul_sharp": 0.17705459892749786, "entropy": 0.5985763669013977} +{"time": "16:14:24", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 12630995, "tok_per_s": 3391.972358669959, "rows_per_s": 4.522810628283892, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5578375458717346, "noul_T": 0.8376916311270307, "choice_ce": 0.6999159455299377, "choice_T": 1.0467250894734454, "score_ce": 1.0730246305465698, "score_T": 0.99517842781877, "noul_auc": 0.7416917333333333, "noul_acc": 0.698, "noul_f1": 0.7772861356932154, "noul_nll": 0.5728344293050468, "noul_ece": 0.03640232239663601, "noul_brier": 0.19540384433301067, "noul_sharp": 0.1810053139925003, "entropy": 0.5957744717597961}, "dropped_long": 0, "peak_gib": 18.28459882736206} diff --git a/e28/S1-E28-E30-e18-uni-syn/DONE b/e28/S1-E28-E30-e18-uni-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..70ce9757dff986dd4634fbca20734734cf009d40 --- /dev/null +++ b/e28/S1-E28-E30-e18-uni-syn/DONE @@ -0,0 +1 @@ +2026-10-21 14:45:06 \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..3e282bdd9e95b466a3a2c3028515bc2e85efc2ac --- /dev/null +++ b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "k_proj", + "o_proj", + "in_proj_qkv", + "up_proj", + "v_proj", + "q_proj", + "in_proj_z", + "gate_proj", + "out_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ca1f9f708a75132f95cffa51c77f4e70ac9186f8 --- /dev/null +++ b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ee5efbe3bb86d6f96c71b5df56cea92a8ecd9352914f343dd83c059bcea7aff +size 160486456 diff --git a/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..17e1b7464a191efbf1ea54a8c53b81f56da87e7f --- /dev/null +++ b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8b2889981ab6064d58b78b203830effb1ddc81d69f3fa481810d37e950a87562 +size 176641832 diff --git a/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..9cad230ab5c5a7f7fdea14a6b69eda6924c97afb --- /dev/null +++ b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.00041243378655053675, + -0.00041269941721111536, + -0.002444965299218893, + -0.0014392428565770388, + -0.00025441867182962596, + 0.002054425422102213, + 0.0015842708526179194, + -0.002470344305038452, + -0.0005563591257669032, + -0.0004927695845253766, + 0.00022808303765486926, + 0.0005532201612368226, + -0.00015229488781187683, + -0.00391890574246645, + -0.004642600659281015, + -0.005127054173499346, + -0.00596853531897068, + -0.0045103468000888824, + -0.004963414743542671, + -0.005049378611147404, + -0.0039865728467702866, + -0.0034670697059482336, + -0.005705875810235739, + -0.0035385722294449806 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/calibration.json b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..b540074d215cb996aea3f57fb6c5e9f11809ab90 --- /dev/null +++ b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.8299663961476075, + "choice": 1.0157569062760596, + "score": 1.0114362819189602 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.8299663961476075, + "ce_before": 0.5577167272567749, + "ce_after": 0.5549060702323914 + }, + "choice": { + "n": 60, + "T": 1.0157569062760596, + "ce_before": 0.6813963055610657, + "ce_after": 0.6813428997993469 + }, + "score": { + "n": 53, + "T": 1.0114362819189602, + "ce_before": 1.070904016494751, + "ce_after": 1.0708658695220947 + } + }, + "source": "models/e28/S1-E28-E30-e18-uni-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/head.safetensors b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1bbb1bd2cf775d2e5bf8f4a9a9d2a8c881439850 --- /dev/null +++ b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe2198db788a4e579d0df6596a66bf5fde42f5ba21726cb058c2c71e6e1d46ca +size 393464 diff --git a/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/judge_config.json b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..15a5c5f11aae5b15c4371298d19df204e95b857c --- /dev/null +++ b/e28/S1-E28-E30-e18-uni-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-E30-e18-uni-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-E30-e18-uni", + "data_manifest": { + "arm": "E28-E30-e18-uni", + "name": null, + "train_sha256": "3cd204c19be775199c17c1e94c0ed211300491e7c14c5a44a26edda56a6fcaee", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 12613706, + "tok_per_s": 3391.4839673053048, + "rows_per_s": 4.528357720408511, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5577167272567749, + "noul_T": 0.8299663961476075, + "choice_ce": 0.6813963055610657, + "choice_T": 1.0157569062760596, + "score_ce": 1.070904016494751, + "score_T": 1.0114362819189602, + "noul_auc": 0.7452885333333332, + "noul_acc": 0.702, + "noul_f1": 0.7766116941529235, + "noul_nll": 0.5715475564301014, + "noul_ece": 0.05453564028441907, + "noul_brier": 0.19476492638229093, + "noul_sharp": 0.17947305738925934, + "entropy": 0.5968093276023865 + }, + "dropped_long": 0, + "peak_gib": 18.284047603607178 + } +} \ No newline at end of file diff --git a/e28/S1-E28-E30-e18-uni-syn/train_log.jsonl b/e28/S1-E28-E30-e18-uni-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..87a07fd1cc729e74f551d98dc9d2e24737ff860a --- /dev/null +++ b/e28/S1-E28-E30-e18-uni-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "13:43:07", "event": "start", "data": "data/e28/E28-E30-e18-uni", "data_manifest": {"arm": "E28-E30-e18-uni", "name": null, "train_sha256": "3cd204c19be775199c17c1e94c0ed211300491e7c14c5a44a26edda56a6fcaee", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.536591291427612} +{"time": "13:44:18", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3466.762124545691, "loss": 0.5949496477842331, "entropy": 0.640225924551487, "sharp": 0.12455910723656416} +{"time": "13:45:29", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3476.6098488276707, "loss": 0.6081669256091118, "entropy": 0.5982899889349937, "sharp": 0.18794278614223003} +{"time": "13:46:43", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3494.653208670433, "loss": 0.6393992714583874, "entropy": 0.6217962205410004, "sharp": 0.16776192374527454} +{"time": "13:47:54", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3497.8789910323317, "loss": 0.5745262280106544, "entropy": 0.5954221896827221, "sharp": 0.18192534055560827} +{"time": "13:49:05", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3501.2120252114933, "loss": 0.5930760502815247, "entropy": 0.5935662686824799, "sharp": 0.1830164734274149} +{"time": "13:50:16", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3507.6699035633656, "loss": 0.5493325293064117, "entropy": 0.5691139921545982, "sharp": 0.19634621124714613} +{"time": "13:51:29", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3512.397311909173, "loss": 0.5660759322345257, "entropy": 0.5459178201854229, "sharp": 0.2223149249330163} +{"time": "13:52:41", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3512.516752848227, "loss": 0.6128996945917606, "entropy": 0.6146528795361519, "sharp": 0.16249231714755297} +{"time": "13:53:53", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3514.613738696456, "loss": 0.589514909312129, "entropy": 0.6076685674488544, "sharp": 0.17052712757140398} +{"time": "13:55:03", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3514.9263032131735, "loss": 0.569058807566762, "entropy": 0.5971218086779118, "sharp": 0.18694508261978626} +{"time": "13:55:22", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.5750044584274292, "noul_T": 1.0238416742940486, "noul_auc": 0.7382322661220906, "noul_acc": 0.696, "noul_f1": 0.7632398753894081, "noul_nll": 0.5802548960670829, "noul_ece": 0.04653878921270371, "noul_brier": 0.19847872596081367, "noul_sharp": 0.20144394040107727, "entropy": 0.570944607257843} +{"time": "13:56:33", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3429.5921395386895, "loss": 0.592361681163311, "entropy": 0.5893814861774445, "sharp": 0.19620598945766687} +{"time": "13:57:45", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3433.5028111232396, "loss": 0.5664022117853165, "entropy": 0.6041635349392891, "sharp": 0.1756181139498949} +{"time": "13:58:56", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3440.737845751685, "loss": 0.569813497364521, "entropy": 0.6072553209960461, "sharp": 0.16806577797979116} +{"time": "14:00:08", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3445.562113663405, "loss": 0.5641683544963598, "entropy": 0.6002166904509068, "sharp": 0.1682357545942068} +{"time": "14:01:17", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3449.3218775841765, "loss": 0.6003596913069487, "entropy": 0.6052957102656364, "sharp": 0.17498073168098927} +{"time": "14:02:28", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3455.451042496693, "loss": 0.6522260792553425, "entropy": 0.5686063058674335, "sharp": 0.18579218816012144} +{"time": "14:03:40", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3460.6389040935906, "loss": 0.5783789865672588, "entropy": 0.5559164136648178, "sharp": 0.2144874269142747} +{"time": "14:04:51", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3464.1086629082038, "loss": 0.5528753027319908, "entropy": 0.6089677587151527, "sharp": 0.17077985126525164} +{"time": "14:06:02", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3465.1802718464933, "loss": 0.6000262424349785, "entropy": 0.5979922190308571, "sharp": 0.174799133092165} +{"time": "14:07:16", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3468.847137344918, "loss": 0.5677642747759819, "entropy": 0.5561921261250973, "sharp": 0.2107423171401024} +{"time": "14:07:36", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5729435086250305, "noul_T": 0.9216740424400314, "noul_auc": 0.7427722738856728, "noul_acc": 0.69, "noul_f1": 0.7611710323574731, "noul_nll": 0.5781940389797091, "noul_ece": 0.04883552676439286, "noul_brier": 0.19744290658552185, "noul_sharp": 0.19219732284545898, "entropy": 0.5864749550819397} +{"time": "14:08:49", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3427.831038942442, "loss": 0.6119560860097408, "entropy": 0.5941413454711437, "sharp": 0.1837636660784483} +{"time": "14:10:00", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3432.32969727582, "loss": 0.6097914557904005, "entropy": 0.6222693584859371, "sharp": 0.16020830813795328} +{"time": "14:11:12", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3436.58228814423, "loss": 0.598186444491148, "entropy": 0.5605175727978349, "sharp": 0.22055327333509922} +{"time": "14:12:26", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3439.9685183748766, "loss": 0.6114900000393391, "entropy": 0.5975478123873472, "sharp": 0.18224158510565758} +{"time": "14:13:37", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3444.3403393581702, "loss": 0.5981363691389561, "entropy": 0.5844681710004807, "sharp": 0.1833025924861431} +{"time": "14:14:49", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3448.449001828537, "loss": 0.5603594332933426, "entropy": 0.5725545324385166, "sharp": 0.20774895325303078} +{"time": "14:15:57", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3450.5548787288426, "loss": 0.5850313156843185, "entropy": 0.5756550170481205, "sharp": 0.19712971337139606} +{"time": "14:17:12", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3453.173682635057, "loss": 0.6226750649511814, "entropy": 0.6029174607247114, "sharp": 0.1816206844523549} +{"time": "14:18:23", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3455.955797744928, "loss": 0.6180335897952318, "entropy": 0.5589930415153503, "sharp": 0.2184185516089201} +{"time": "14:19:34", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3457.8983749581244, "loss": 0.6113746743649244, "entropy": 0.590357780456543, "sharp": 0.18336332216858864} +{"time": "14:19:54", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5747140645980835, "noul_T": 0.793686717643164, "noul_auc": 0.7480548851496178, "noul_acc": 0.698, "noul_f1": 0.765891472868217, "noul_nll": 0.5772964083105325, "noul_ece": 0.07298361307382584, "noul_brier": 0.19686431033055926, "noul_sharp": 0.16942602396011353, "entropy": 0.6068527698516846} +{"time": "14:21:05", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3430.3135611502767, "loss": 0.6120510920882225, "entropy": 0.6184627562761307, "sharp": 0.1665957486256957} +{"time": "14:22:18", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3432.992128799039, "loss": 0.6016992554068565, "entropy": 0.6020928770303726, "sharp": 0.16728418786078691} +{"time": "14:23:28", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3436.3307905271927, "loss": 0.5902691222727299, "entropy": 0.5414585061371326, "sharp": 0.2378454441204667} +{"time": "14:24:40", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3438.8656869516535, "loss": 0.6034007780253887, "entropy": 0.5846098959445953, "sharp": 0.1929677538573742} +{"time": "14:25:52", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3441.199988959473, "loss": 0.6179255694150925, "entropy": 0.6081629693508148, "sharp": 0.16352617554366589} +{"time": "14:27:05", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3443.6312918181648, "loss": 0.6538338474929333, "entropy": 0.6181873865425587, "sharp": 0.15229598060250282} +{"time": "14:28:17", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3446.009040123888, "loss": 0.589688265696168, "entropy": 0.5980271697044373, "sharp": 0.188619633205235} +{"time": "14:29:28", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3447.6589989591484, "loss": 0.6557577103376389, "entropy": 0.6338713206350803, "sharp": 0.14343180041760206} +{"time": "14:30:39", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3450.634073814165, "loss": 0.6004322469234467, "entropy": 0.5726072564721107, "sharp": 0.19849284552037716} +{"time": "14:31:52", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3452.8432226983077, "loss": 0.5903890579938889, "entropy": 0.5854857377707958, "sharp": 0.19801112823188305} +{"time": "14:32:12", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5733904838562012, "noul_T": 0.8492874211709703, "noul_auc": 0.7497088656731532, "noul_acc": 0.704, "noul_f1": 0.7694704049844237, "noul_nll": 0.5749362139180303, "noul_ece": 0.05253567343950273, "noul_brier": 0.1959223576056529, "noul_sharp": 0.1782318502664566, "entropy": 0.5977874398231506} +{"time": "14:33:23", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3432.396443890471, "loss": 0.6125591844320297, "entropy": 0.6213195100426674, "sharp": 0.15995072573423386} +{"time": "14:34:35", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3434.3791165512043, "loss": 0.5823782123625278, "entropy": 0.5569762773811817, "sharp": 0.2158653885126114} +{"time": "14:35:47", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3436.4916115297133, "loss": 0.6104585621505976, "entropy": 0.5876182718202472, "sharp": 0.18281534127891064} +{"time": "14:36:56", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3438.4522694651478, "loss": 0.6208794340491295, "entropy": 0.6022915542125702, "sharp": 0.1756770983338356} +{"time": "14:38:09", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3440.060883799379, "loss": 0.5625946559011936, "entropy": 0.5846921280026436, "sharp": 0.186237420886755} +{"time": "14:39:18", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3442.839379699653, "loss": 0.5980923995375633, "entropy": 0.5843414142727852, "sharp": 0.1979096159338951} +{"time": "14:40:28", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3444.561364158745, "loss": 0.6685153692960739, "entropy": 0.6140761710703373, "sharp": 0.1611163830384612} +{"time": "14:41:40", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3446.1591184651006, "loss": 0.597884438931942, "entropy": 0.5801304578781128, "sharp": 0.1984165571630001} +{"time": "14:42:52", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3448.360332114174, "loss": 0.5433237329125404, "entropy": 0.568408764898777, "sharp": 0.20580919086933136} +{"time": "14:44:03", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3449.3188160428435, "loss": 0.5883278921246529, "entropy": 0.5875190831720829, "sharp": 0.1883558090776205} +{"time": "14:44:23", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5717165470123291, "noul_T": 0.8290715518200243, "noul_auc": 0.7506539974008878, "noul_acc": 0.7, "noul_f1": 0.7663551401869159, "noul_nll": 0.5738489653915166, "noul_ece": 0.05715590554475785, "noul_brier": 0.19556824781570067, "noul_sharp": 0.17651602625846863, "entropy": 0.5993005037307739} +{"time": "14:45:06", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 12613706, "tok_per_s": 3391.4839673053048, "rows_per_s": 4.528357720408511, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5577167272567749, "noul_T": 0.8299663961476075, "choice_ce": 0.6813963055610657, "choice_T": 1.0157569062760596, "score_ce": 1.070904016494751, "score_T": 1.0114362819189602, "noul_auc": 0.7452885333333332, "noul_acc": 0.702, "noul_f1": 0.7766116941529235, "noul_nll": 0.5715475564301014, "noul_ece": 0.05453564028441907, "noul_brier": 0.19476492638229093, "noul_sharp": 0.17947305738925934, "entropy": 0.5968093276023865}, "dropped_long": 0, "peak_gib": 18.284047603607178} diff --git a/e28/S1-E28-E30-mix-rw-syn/DONE b/e28/S1-E28-E30-mix-rw-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..698aa7b67d8b58f7031b8ff09cf43b95c4a4828f --- /dev/null +++ b/e28/S1-E28-E30-mix-rw-syn/DONE @@ -0,0 +1 @@ +2026-10-21 13:16:09 \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..0fe1b16fbc2fe79fd738d89df821d706596522f5 --- /dev/null +++ b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "q_proj", + "k_proj", + "o_proj", + "in_proj_qkv", + "up_proj", + "gate_proj", + "down_proj", + "out_proj", + "in_proj_z", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9729a1400e1eca06a862657cc40c2a2ced2c9997 --- /dev/null +++ b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9bd674aa272f30369bcb23835a47849d0537a195aa025ef419fbc41575798887 +size 160486456 diff --git a/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a3217d796b0262a1ec920658f12792e0c31ea753 --- /dev/null +++ b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6487e9f15bfc3fdc07115f8f4d9e77c52e87e78f6e9cc29ba603c5fdbdcfc143 +size 176641832 diff --git a/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..69cd4a03acf2b57caa8c2774ee8329ce2fefa592 --- /dev/null +++ b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.00042491714702919126, + -0.00042517148540355265, + -0.0027090017683804035, + -0.001921229064464569, + -0.0007290767389349639, + 0.001882426324300468, + 0.0014677918516099453, + -0.0027623651549220085, + -0.000474259490147233, + -0.0004921925137750804, + -0.00029579506372101605, + 0.00034974361187778413, + -8.164704195223749e-05, + -0.0034557334147393703, + -0.004367884714156389, + -0.005152189638465643, + -0.0059998719953000546, + -0.00439545139670372, + -0.00494065647944808, + -0.004992631729692221, + -0.003735068952664733, + -0.0033735833130776882, + -0.005707588978111744, + -0.0034806577023118734 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/calibration.json b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..33a18658e5f747bf0f34299ad0cc0995de1b40c7 --- /dev/null +++ b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.7641089818788265, + "choice": 0.9619936000838222, + "score": 0.9497450247643356 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.7641089818788265, + "ce_before": 0.5750262141227722, + "ce_after": 0.5698644518852234 + }, + "choice": { + "n": 60, + "T": 0.9619936000838222, + "ce_before": 0.6642542481422424, + "ce_after": 0.6639342904090881 + }, + "score": { + "n": 53, + "T": 0.9497450247643356, + "ce_before": 1.0545263290405273, + "ce_after": 1.0537408590316772 + } + }, + "source": "models/e28/S1-E28-E30-mix-rw-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/head.safetensors b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7b7110c5f0eee815ca4c69a5e25df84d68206744 --- /dev/null +++ b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd894b625cc78f0a99e0662cbeede040fed2f10f4ffe720ea3b4c24907e6663f +size 393464 diff --git a/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/judge_config.json b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..b485001d0294a16da5b1bf588e55cbbeec413016 --- /dev/null +++ b/e28/S1-E28-E30-mix-rw-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-E30-mix-rw-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-E30-mix-rw", + "data_manifest": { + "arm": "E28-E30-mix-rw", + "name": null, + "train_sha256": "66b45988eaa25b98de07dad009b734e0e6b2e1954e0ef609f92ad4eafd234b29", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9858307, + "tok_per_s": 3322.4924505729546, + "rows_per_s": 5.676169124838469, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5750262141227722, + "noul_T": 0.7641089818788265, + "choice_ce": 0.6642542481422424, + "choice_T": 0.9619936000838222, + "score_ce": 1.0545263290405273, + "score_T": 0.9497450247643356, + "noul_auc": 0.7218901333333334, + "noul_acc": 0.692, + "noul_f1": 0.7780979827089337, + "noul_nll": 0.5908899419978261, + "noul_ece": 0.05135784411430359, + "noul_brier": 0.2021236720863354, + "noul_sharp": 0.15754856169223785, + "entropy": 0.6208320260047913 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-rw-syn/train_log.jsonl b/e28/S1-E28-E30-mix-rw-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..ef9c73030c84766dca41f9d78e9aa42c473b51ec --- /dev/null +++ b/e28/S1-E28-E30-mix-rw-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "12:26:41", "event": "start", "data": "data/e28/E28-E30-mix-rw", "data_manifest": {"arm": "E28-E30-mix-rw", "name": null, "train_sha256": "66b45988eaa25b98de07dad009b734e0e6b2e1954e0ef609f92ad4eafd234b29", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.816522836685181} +{"time": "12:27:39", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3398.5104450242075, "loss": 0.6617981046438217, "entropy": 0.6758278906345367, "sharp": 0.07641357369720936} +{"time": "12:28:36", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3446.3254871859986, "loss": 0.5870574302971363, "entropy": 0.5065327882766724, "sharp": 0.27059395238757133} +{"time": "12:29:32", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3448.3366269948433, "loss": 0.6643475610762835, "entropy": 0.6517691090703011, "sharp": 0.12636451981961727} +{"time": "12:30:29", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3450.1062169909283, "loss": 0.6300797145813704, "entropy": 0.661946514621377, "sharp": 0.09981008898466825} +{"time": "12:31:23", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3460.7687705118105, "loss": 0.6509177535772324, "entropy": 0.6351496540009975, "sharp": 0.15960458107292652} +{"time": "12:32:20", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3470.1564362123922, "loss": 0.5559498108923435, "entropy": 0.6086036339402199, "sharp": 0.17464126646518707} +{"time": "12:33:17", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3474.2649007262744, "loss": 0.5867289453744888, "entropy": 0.6473392620682716, "sharp": 0.12959711253643036} +{"time": "12:34:12", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3479.233186561942, "loss": 0.6344430595636368, "entropy": 0.6369583383202553, "sharp": 0.14375981222838163} +{"time": "12:35:09", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3483.3830758298523, "loss": 0.6572696678340435, "entropy": 0.5704085677862167, "sharp": 0.22912083752453327} +{"time": "12:36:07", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3484.210377810486, "loss": 0.5975021179765463, "entropy": 0.6458938233554363, "sharp": 0.13581185787916183} +{"time": "12:36:27", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.5912001132965088, "noul_T": 0.749338714212629, "noul_auc": 0.7274645153668292, "noul_acc": 0.704, "noul_f1": 0.7817109144542773, "noul_nll": 0.5935823530256749, "noul_ece": 0.07118012642860415, "noul_brier": 0.2026182278811666, "noul_sharp": 0.15321074426174164, "entropy": 0.6266841888427734} +{"time": "12:37:23", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3376.441763408906, "loss": 0.6423913687467575, "entropy": 0.6464557480067015, "sharp": 0.12764806021004915} +{"time": "12:38:18", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3381.8700606511748, "loss": 0.6129378750920296, "entropy": 0.6169216334819794, "sharp": 0.16408988367766142} +{"time": "12:39:14", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3391.601101806736, "loss": 0.6241332367062569, "entropy": 0.6356232985854149, "sharp": 0.13230206351727247} +{"time": "12:40:11", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3397.221062238907, "loss": 0.6194629743695259, "entropy": 0.5942576006054878, "sharp": 0.184342703782022} +{"time": "12:41:09", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3403.856411952408, "loss": 0.6410548184067011, "entropy": 0.6603901982307434, "sharp": 0.10944096837192774} +{"time": "12:42:06", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3411.828988671562, "loss": 0.6618923619389534, "entropy": 0.5877530016005039, "sharp": 0.18872639536857605} +{"time": "12:43:01", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3419.0855820924553, "loss": 0.6553576812148094, "entropy": 0.6332240253686905, "sharp": 0.15175582747906446} +{"time": "12:43:57", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3423.524774992223, "loss": 0.6165152825415134, "entropy": 0.6027723997831345, "sharp": 0.19378135353326797} +{"time": "12:44:55", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3427.2603413583615, "loss": 0.6480486653745174, "entropy": 0.630643755197525, "sharp": 0.1512931054458022} +{"time": "12:45:51", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3430.463135146777, "loss": 0.5993139799684286, "entropy": 0.6381316836923361, "sharp": 0.14204187504947186} +{"time": "12:46:11", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5899976491928101, "noul_T": 0.6841149011909993, "noul_auc": 0.7421140571467147, "noul_acc": 0.706, "noul_f1": 0.7776096822995462, "noul_nll": 0.5895531940460205, "noul_ece": 0.06612831807136535, "noul_brier": 0.20107843339349146, "noul_sharp": 0.14440566301345825, "entropy": 0.6332550644874573} +{"time": "12:47:07", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3376.10929641793, "loss": 0.6361674629151821, "entropy": 0.5290279015898705, "sharp": 0.2586303371936083} +{"time": "12:48:03", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3382.043840529214, "loss": 0.6442053057253361, "entropy": 0.6576089113950729, "sharp": 0.11792709305882454} +{"time": "12:49:00", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3386.8080548486837, "loss": 0.5984862688928843, "entropy": 0.5767329707741737, "sharp": 0.222583731636405} +{"time": "12:49:56", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3389.7393925994, "loss": 0.7062062099575996, "entropy": 0.6577959135174751, "sharp": 0.10799758229404688} +{"time": "12:50:51", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3391.237543120939, "loss": 0.6199326664209366, "entropy": 0.6112727038562298, "sharp": 0.18155966233462095} +{"time": "12:51:48", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3395.5510864855173, "loss": 0.623970391228795, "entropy": 0.6604787595570087, "sharp": 0.10022117104381323} +{"time": "12:52:45", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3399.0224133251268, "loss": 0.6235006228089333, "entropy": 0.568578913807869, "sharp": 0.21520205028355122} +{"time": "12:53:42", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3402.5215593303838, "loss": 0.5979901999235153, "entropy": 0.6021059677004814, "sharp": 0.18195396661758423} +{"time": "12:54:38", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3405.2805455272483, "loss": 0.6152922660112381, "entropy": 0.6330707669258118, "sharp": 0.1516977259889245} +{"time": "12:55:34", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3409.06461184697, "loss": 0.5994616076350212, "entropy": 0.6110654771327972, "sharp": 0.17343833576887846} +{"time": "12:55:54", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5856232047080994, "noul_T": 0.8958703537984413, "noul_auc": 0.7295404296973892, "noul_acc": 0.706, "noul_f1": 0.7815750371471025, "noul_nll": 0.5901440802961588, "noul_ece": 0.05012203615903854, "noul_brier": 0.2013562340443956, "noul_sharp": 0.17469966411590576, "entropy": 0.6016364693641663} +{"time": "12:56:52", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3373.2022725328675, "loss": 0.6347501836717129, "entropy": 0.6354512274265289, "sharp": 0.13818068150430918} +{"time": "12:57:48", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3375.5885970252652, "loss": 0.5916188042610884, "entropy": 0.5902127977460623, "sharp": 0.19401217252016068} +{"time": "12:58:45", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3378.68280480534, "loss": 0.6172374859452248, "entropy": 0.6208569929003716, "sharp": 0.15783602558076382} +{"time": "12:59:41", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3381.138089681666, "loss": 0.6336306110024452, "entropy": 0.6333046332001686, "sharp": 0.15671599842607975} +{"time": "13:00:38", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3384.998292931626, "loss": 0.6413889862596989, "entropy": 0.6416175588965416, "sharp": 0.13174898084253073} +{"time": "13:01:35", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3388.5660794744135, "loss": 0.6692870035767555, "entropy": 0.6100635975599289, "sharp": 0.175360394641757} +{"time": "13:02:30", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3390.894423687449, "loss": 0.5844628922641277, "entropy": 0.6072426289319992, "sharp": 0.1719667688012123} +{"time": "13:03:25", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3393.2922594710753, "loss": 0.6599180158227682, "entropy": 0.6330446414649487, "sharp": 0.14461122080683708} +{"time": "13:04:24", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3395.5798670332797, "loss": 0.6219851486384869, "entropy": 0.633172407746315, "sharp": 0.1486526671797037} +{"time": "13:05:18", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3398.124065812128, "loss": 0.6529631316661835, "entropy": 0.621990691870451, "sharp": 0.16150179784744978} +{"time": "13:05:38", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5857885479927063, "noul_T": 0.7984685586236674, "noul_auc": 0.736173229143812, "noul_acc": 0.692, "noul_f1": 0.7735294117647059, "noul_nll": 0.5880960110425949, "noul_ece": 0.05230468237400056, "noul_brier": 0.20078692976809265, "noul_sharp": 0.16155387461185455, "entropy": 0.6157422065734863} +{"time": "13:06:35", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3371.623488411751, "loss": 0.643172562122345, "entropy": 0.6340844258666039, "sharp": 0.14288697391748428} +{"time": "13:07:32", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3374.892286346895, "loss": 0.6080732196569443, "entropy": 0.6013886108994484, "sharp": 0.18995149992406368} +{"time": "13:08:30", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3378.358031928968, "loss": 0.6302251946181059, "entropy": 0.6078975051641464, "sharp": 0.1825722698122263} +{"time": "13:09:28", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3381.1063759808644, "loss": 0.6565866321325302, "entropy": 0.648265577852726, "sharp": 0.11372941732406616} +{"time": "13:10:26", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3383.852192269609, "loss": 0.6076048277318478, "entropy": 0.5934285633265972, "sharp": 0.1934272302314639} +{"time": "13:11:20", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3385.1301214308232, "loss": 0.5958774704486132, "entropy": 0.5815382711589336, "sharp": 0.21349548548460007} +{"time": "13:12:16", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3386.8454491660773, "loss": 0.6104066446423531, "entropy": 0.6062584742903709, "sharp": 0.17242817766964436} +{"time": "13:13:12", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3389.093656927826, "loss": 0.654010783880949, "entropy": 0.5944750271737576, "sharp": 0.19629749655723572} +{"time": "13:14:09", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3392.1860627493115, "loss": 0.6189305819571018, "entropy": 0.619609959423542, "sharp": 0.16488080006092787} +{"time": "13:15:06", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3393.8455932985958, "loss": 0.6065668482333422, "entropy": 0.6191403977572918, "sharp": 0.16083800420165062} +{"time": "13:15:26", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5894796252250671, "noul_T": 0.755261746732893, "noul_auc": 0.7350762012455486, "noul_acc": 0.708, "noul_f1": 0.781437125748503, "noul_nll": 0.5907158334106207, "noul_ece": 0.08182558190822603, "noul_brier": 0.20162164802819404, "noul_sharp": 0.15212295949459076, "entropy": 0.6238287687301636} +{"time": "13:16:09", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9858307, "tok_per_s": 3322.4924505729546, "rows_per_s": 5.676169124838469, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5750262141227722, "noul_T": 0.7641089818788265, "choice_ce": 0.6642542481422424, "choice_T": 0.9619936000838222, "score_ce": 1.0545263290405273, "score_T": 0.9497450247643356, "noul_auc": 0.7218901333333334, "noul_acc": 0.692, "noul_f1": 0.7780979827089337, "noul_nll": 0.5908899419978261, "noul_ece": 0.05135784411430359, "noul_brier": 0.2021236720863354, "noul_sharp": 0.15754856169223785, "entropy": 0.6208320260047913}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-E30-mix-uni-syn/DONE b/e28/S1-E28-E30-mix-uni-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..1506575ee1803d27836e666074450ea78b505f76 --- /dev/null +++ b/e28/S1-E28-E30-mix-uni-syn/DONE @@ -0,0 +1 @@ +2026-10-21 11:59:57 \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..10f7a91e8c08ae9aec21e701f48dcd3b1ec6ebf5 --- /dev/null +++ b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "v_proj", + "in_proj_z", + "gate_proj", + "out_proj", + "k_proj", + "q_proj", + "in_proj_qkv", + "up_proj", + "down_proj", + "o_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f35a35e36c0791c4589766fb856dbc28e7aefa24 --- /dev/null +++ b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:52afb1fae187be03c2748f9216c649521c3f63967af75eb92f47b9eafc872e84 +size 160486456 diff --git a/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..080fc93f69633d905d9a2e6ac2c9c0acf389564f --- /dev/null +++ b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e29934502745e9ad8999b8ff967aa206d487c0d8268b9c6f277640e675aed47 +size 176641832 diff --git a/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..33fbe61edede12c7e674d371d53c0ccf1921826f --- /dev/null +++ b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.00038413761649280787, + -0.00038439081981778145, + -0.002247646450996399, + -0.0015133899869397283, + -0.0006128472159616649, + 0.0021903254091739655, + 0.0014095399528741837, + -0.0026955355424433947, + -0.0002967127365991473, + -0.0007888193940743804, + 0.000303100619930774, + 0.000279925880022347, + -0.000601482461206615, + -0.0034952836576849222, + -0.004261689260601997, + -0.005193612072616816, + -0.005949186161160469, + -0.004410287830978632, + -0.004946717526763678, + -0.005019421223551035, + -0.0037548185791820288, + -0.003413393860682845, + -0.005750156007707119, + -0.0034985444508492947 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/calibration.json b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..5b588adb8f90ef854706eb2b14f5152e5d7b850d --- /dev/null +++ b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.8415497215915553, + "choice": 1.0720952112642603, + "score": 0.9580860445008661 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.8415497215915553, + "ce_before": 0.5767975449562073, + "ce_after": 0.5745895504951477 + }, + "choice": { + "n": 60, + "T": 1.0720952112642603, + "ce_before": 0.6827930212020874, + "ce_after": 0.6816638708114624 + }, + "score": { + "n": 53, + "T": 0.9580860445008661, + "ce_before": 1.051763892173767, + "ce_after": 1.0512245893478394 + } + }, + "source": "models/e28/S1-E28-E30-mix-uni-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/head.safetensors b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5b1fbdeffdcd6d15c07165971e1f19065eefcb97 --- /dev/null +++ b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ba9eb3230c6f75d4a1f6fd5187e04c8da74d49420152dbba208198fbcd44391e +size 393464 diff --git a/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/judge_config.json b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..8e6dce707f2fa4b1e80f39c5a112d06ffc52db3b --- /dev/null +++ b/e28/S1-E28-E30-mix-uni-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-E30-mix-uni-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-E30-mix-uni", + "data_manifest": { + "arm": "E28-E30-mix-uni", + "name": null, + "train_sha256": "66c6a065117057df65d08b2b307a32a3e3155348a10465f31e20c719819c6339", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9786193, + "tok_per_s": 3315.1212953546064, + "rows_per_s": 5.7053108215622235, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5767975449562073, + "noul_T": 0.8415497215915553, + "choice_ce": 0.6827930212020874, + "choice_T": 1.0720952112642603, + "score_ce": 1.051763892173767, + "score_T": 0.9580860445008661, + "noul_auc": 0.7142656000000001, + "noul_acc": 0.689, + "noul_f1": 0.7751265365148229, + "noul_nll": 0.5924707644954323, + "noul_ece": 0.050650765240192415, + "noul_brier": 0.20300665197023557, + "noul_sharp": 0.16800683736801147, + "entropy": 0.6124101281166077 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-E30-mix-uni-syn/train_log.jsonl b/e28/S1-E28-E30-mix-uni-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..431f2e1120e2ec7633bd7c99bda5215ae320316c --- /dev/null +++ b/e28/S1-E28-E30-mix-uni-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "11:10:45", "event": "start", "data": "data/e28/E28-E30-mix-uni", "data_manifest": {"arm": "E28-E30-mix-uni", "name": null, "train_sha256": "66c6a065117057df65d08b2b307a32a3e3155348a10465f31e20c719819c6339", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.846560955047607} +{"time": "11:11:41", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3443.8877501593747, "loss": 0.6262777708470821, "entropy": 0.6350177302956581, "sharp": 0.14207495655864477} +{"time": "11:12:37", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3448.8760442278954, "loss": 0.6659292206168175, "entropy": 0.6351186037063599, "sharp": 0.1585685033351183} +{"time": "11:13:33", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3468.6852368187187, "loss": 0.7106882184743881, "entropy": 0.6595125049352646, "sharp": 0.10771670192480087} +{"time": "11:14:28", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3461.892704914454, "loss": 0.6871962323784828, "entropy": 0.6570172794163227, "sharp": 0.10243075247853994} +{"time": "11:15:25", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3471.5377607664727, "loss": 0.6552312038838863, "entropy": 0.667052298784256, "sharp": 0.09706676099449396} +{"time": "11:16:20", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3470.407184601818, "loss": 0.6605521105229855, "entropy": 0.6164332404732704, "sharp": 0.16601129341870546} +{"time": "11:17:17", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3473.1577656669656, "loss": 0.6395523957908154, "entropy": 0.6380520388484001, "sharp": 0.1491299606859684} +{"time": "11:18:13", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3477.3957317463505, "loss": 0.6371722742915154, "entropy": 0.6528833620250225, "sharp": 0.121398008428514} +{"time": "11:19:09", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3480.6446396226347, "loss": 0.6607006564736366, "entropy": 0.6257094517350197, "sharp": 0.1571742706000805} +{"time": "11:20:06", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3483.6069992531707, "loss": 0.6417962200939655, "entropy": 0.6607039645314217, "sharp": 0.10791026055812836} +{"time": "11:20:26", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.6048566102981567, "noul_T": 0.6141053135000148, "noul_auc": 0.7323589475283117, "noul_acc": 0.716, "noul_f1": 0.7767295597484277, "noul_nll": 0.6021790899336338, "noul_ece": 0.10803234994411472, "noul_brier": 0.2066901547291324, "noul_sharp": 0.12016980350017548, "entropy": 0.6481845378875732} +{"time": "11:21:23", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3377.3697396451357, "loss": 0.691669262945652, "entropy": 0.6485184952616692, "sharp": 0.126510726287961} +{"time": "11:22:20", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3386.6464675467478, "loss": 0.6227710172533989, "entropy": 0.6643407791852951, "sharp": 0.0989102590829134} +{"time": "11:23:15", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3392.10279527126, "loss": 0.6345570869743824, "entropy": 0.6353756655007601, "sharp": 0.14184112381190062} +{"time": "11:24:12", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3400.25700274714, "loss": 0.6644854545593262, "entropy": 0.6568174585700035, "sharp": 0.10916892066597939} +{"time": "11:25:09", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3407.5577297441937, "loss": 0.6628513261675835, "entropy": 0.6120313704013824, "sharp": 0.17041109316051006} +{"time": "11:26:08", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3411.8707335461327, "loss": 0.7355445679277182, "entropy": 0.6772512830793858, "sharp": 0.06662071496248245} +{"time": "11:27:04", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3416.5427804206265, "loss": 0.6487440392374992, "entropy": 0.627815917134285, "sharp": 0.1477979915216565} +{"time": "11:28:00", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3418.91022541653, "loss": 0.6276426836848259, "entropy": 0.6544394418597221, "sharp": 0.10995172150433064} +{"time": "11:28:56", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3422.3492025976284, "loss": 0.6820836253464222, "entropy": 0.6257366873323917, "sharp": 0.1452926304191351} +{"time": "11:29:52", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3424.239775066111, "loss": 0.6783181801438332, "entropy": 0.6620986387133598, "sharp": 0.10394556634128094} +{"time": "11:30:12", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.6135072112083435, "noul_T": 0.7345515158131158, "noul_auc": 0.7341141921655332, "noul_acc": 0.684, "noul_f1": 0.7285223367697594, "noul_nll": 0.6100223219394684, "noul_ece": 0.1148088651895523, "noul_brier": 0.2101808475459673, "noul_sharp": 0.13129998743534088, "entropy": 0.6429041624069214} +{"time": "11:31:06", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3371.613742219997, "loss": 0.655721016228199, "entropy": 0.6506842449307442, "sharp": 0.12078169919550419} +{"time": "11:32:04", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3376.731063483007, "loss": 0.6074702683836222, "entropy": 0.6286280266940594, "sharp": 0.1483098240569234} +{"time": "11:33:00", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3382.3818544728633, "loss": 0.6853180378675461, "entropy": 0.6122780665755272, "sharp": 0.16668810602277517} +{"time": "11:33:55", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3386.510665521531, "loss": 0.6349831782281399, "entropy": 0.6374031007289886, "sharp": 0.13709043338894844} +{"time": "11:34:52", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3389.84440425201, "loss": 0.6617998257279396, "entropy": 0.6560108922421932, "sharp": 0.11490273009985685} +{"time": "11:35:47", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3393.043934590687, "loss": 0.6786966770887375, "entropy": 0.6296423226594925, "sharp": 0.15470370650291443} +{"time": "11:36:42", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3395.307145594965, "loss": 0.6309544369578362, "entropy": 0.6555880084633827, "sharp": 0.11157542932778597} +{"time": "11:37:37", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3398.9195725536656, "loss": 0.6893973089754581, "entropy": 0.6453859508037567, "sharp": 0.12598618026822805} +{"time": "11:38:35", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3401.4098528059153, "loss": 0.7022407278418541, "entropy": 0.654004905372858, "sharp": 0.11092838738113642} +{"time": "11:39:31", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3403.4814078849595, "loss": 0.7073084115982056, "entropy": 0.6600758656859398, "sharp": 0.10507288295775652} +{"time": "11:39:51", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5957980155944824, "noul_T": 0.7354819437582658, "noul_auc": 0.7239202713878247, "noul_acc": 0.696, "noul_f1": 0.7738095238095238, "noul_nll": 0.5982052974402905, "noul_ece": 0.0704976155757904, "noul_brier": 0.20519243117207048, "noul_sharp": 0.14060336351394653, "entropy": 0.6293926239013672} +{"time": "11:40:48", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3368.629345009118, "loss": 0.6345702018588781, "entropy": 0.6467453800141811, "sharp": 0.12419570051133633} +{"time": "11:41:43", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3370.665215848845, "loss": 0.6466725468635559, "entropy": 0.6435500085353851, "sharp": 0.1302737034857273} +{"time": "11:42:39", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3373.3478851477585, "loss": 0.6761657521128654, "entropy": 0.6601769179105759, "sharp": 0.10026721656322479} +{"time": "11:43:36", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3376.9184153932088, "loss": 0.655341137200594, "entropy": 0.6611673422157764, "sharp": 0.10098677314817905} +{"time": "11:44:32", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3379.2817996572403, "loss": 0.6435482576489449, "entropy": 0.6001698710024357, "sharp": 0.18823308032006025} +{"time": "11:45:27", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3381.884048518836, "loss": 0.6803960874676704, "entropy": 0.6400693878531456, "sharp": 0.13950764760375023} +{"time": "11:46:24", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3384.241389852097, "loss": 0.6741290576756, "entropy": 0.6435752473771572, "sharp": 0.12162770796567202} +{"time": "11:47:19", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3386.9792562414864, "loss": 0.6685706060379744, "entropy": 0.6324643157422543, "sharp": 0.14916173461824656} +{"time": "11:48:14", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3388.5346663024, "loss": 0.7060552649199963, "entropy": 0.6424734331667423, "sharp": 0.13531916216015816} +{"time": "11:49:10", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3391.557729161689, "loss": 0.6456914488226175, "entropy": 0.6640833299607038, "sharp": 0.09684538189321756} +{"time": "11:49:30", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5968390107154846, "noul_T": 0.6773519822206349, "noul_auc": 0.7319201363690064, "noul_acc": 0.692, "noul_f1": 0.772189349112426, "noul_nll": 0.5963751047253608, "noul_ece": 0.07572836101055147, "noul_brier": 0.20397312116312713, "noul_sharp": 0.1381661593914032, "entropy": 0.6379590034484863} +{"time": "11:50:26", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3365.269244714983, "loss": 0.6728063300251961, "entropy": 0.6604369506239891, "sharp": 0.10716191958636045} +{"time": "11:51:24", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3368.6123461944035, "loss": 0.6814468000084162, "entropy": 0.6591662727296352, "sharp": 0.11162697896361351} +{"time": "11:52:19", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3370.8533619383893, "loss": 0.6765133682638407, "entropy": 0.6549635753035545, "sharp": 0.10235160775482655} +{"time": "11:53:15", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3373.740250970939, "loss": 0.6967063751071692, "entropy": 0.6452214270830154, "sharp": 0.12953200284391642} +{"time": "11:54:13", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3376.1797913249534, "loss": 0.648522550240159, "entropy": 0.6530167013406754, "sharp": 0.11750649195164442} +{"time": "11:55:09", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3378.4287922697936, "loss": 0.6533721350133419, "entropy": 0.6649773940443993, "sharp": 0.0958434734493494} +{"time": "11:56:06", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3380.8357376646263, "loss": 0.691974438726902, "entropy": 0.6363588199019432, "sharp": 0.1376436948776245} +{"time": "11:57:03", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3382.8555923720182, "loss": 0.6520401109009981, "entropy": 0.6657268181443214, "sharp": 0.09253840893507004} +{"time": "11:57:58", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3384.3489679525537, "loss": 0.6411418356001377, "entropy": 0.6559964716434479, "sharp": 0.11062734015285969} +{"time": "11:58:54", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3386.623301800667, "loss": 0.6552579626441002, "entropy": 0.6548320967704058, "sharp": 0.11682656779885292} +{"time": "11:59:14", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5899556279182434, "noul_T": 0.8093489071042997, "noul_auc": 0.7310087593458339, "noul_acc": 0.696, "noul_f1": 0.7717717717717718, "noul_nll": 0.5889124748259783, "noul_ece": 0.060712868213653584, "noul_brier": 0.20090344300966484, "noul_sharp": 0.16099826991558075, "entropy": 0.6177083253860474} +{"time": "11:59:57", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9786193, "tok_per_s": 3315.1212953546064, "rows_per_s": 5.7053108215622235, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5767975449562073, "noul_T": 0.8415497215915553, "choice_ce": 0.6827930212020874, "choice_T": 1.0720952112642603, "score_ce": 1.051763892173767, "score_T": 0.9580860445008661, "noul_auc": 0.7142656000000001, "noul_acc": 0.689, "noul_f1": 0.7751265365148229, "noul_nll": 0.5924707644954323, "noul_ece": 0.050650765240192415, "noul_brier": 0.20300665197023557, "noul_sharp": 0.16800683736801147, "entropy": 0.6124101281166077}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-I4-n16k-syn/DONE b/e28/S1-E28-I4-n16k-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..3671ade6d85847e07c262fc484ce42d1410a0a82 --- /dev/null +++ b/e28/S1-E28-I4-n16k-syn/DONE @@ -0,0 +1 @@ +2026-10-21 04:29:09 \ No newline at end of file diff --git a/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..1c9b228d39df69992b2484c399f557a4caa7c554 --- /dev/null +++ b/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "q_proj", + "o_proj", + "k_proj", + "v_proj", + "out_proj", + "up_proj", + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e49af0f6b136f02cd092225dd24eaeaf231531e7 --- /dev/null +++ b/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b40f7b41aa2ee02d66f40d0e3bee0cd1fdb70477148cc1fd66e6340976ac1e60 +size 160486456 diff --git a/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a19a98dd5600b6194d70d5a27dd050fc1bae15dc --- /dev/null +++ b/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0ff5866d04554111c4845a3a5dcd99d545e593ed63fa94e65180653d404365c3 +size 176641832 diff --git a/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..da689e839f9bdddadb12e2d9a44c367c6170893d --- /dev/null +++ b/e28/S1-E28-I4-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.0004493747837841511, + -0.00044962650281377137, + -0.002473356667906046, + -0.001983140828087926, + -0.0006738185184076428, + 0.0016270865453407168, + 0.001159157371148467, + -0.00242731929756701, + -0.00042943868902511895, + -0.0004960581427440047, + 0.0002525956660974771, + 9.827975736698136e-05, + -0.00019377579155843705, + -0.0034460132010281086, + -0.004351620562374592, + -0.005213302094489336, + -0.006012967322021723, + -0.0044012414291501045, + -0.00493605574592948, + -0.004993426147848368, + -0.0037619464565068483, + -0.00337326992303133, + -0.0056798700243234634, + -0.0034921530168503523 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-I4-n16k-syn/ckpt-16000/calibration.json b/e28/S1-E28-I4-n16k-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..123e99d77bd814a74bf0d1b059cb085f61072fbd --- /dev/null +++ b/e28/S1-E28-I4-n16k-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.6919772888709426, + "choice": 1.0109731528245147, + "score": 0.9875175650796282 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.6919772888709426, + "ce_before": 0.5806214809417725, + "ce_after": 0.5719630122184753 + }, + "choice": { + "n": 60, + "T": 1.0109731528245147, + "ce_before": 0.6720502376556396, + "ce_after": 0.6720252633094788 + }, + "score": { + "n": 53, + "T": 0.9875175650796282, + "ce_before": 1.061937928199768, + "ce_after": 1.0618877410888672 + } + }, + "source": "models/e28/S1-E28-I4-n16k-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-I4-n16k-syn/ckpt-16000/head.safetensors b/e28/S1-E28-I4-n16k-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..308066b57bab8018c140ebf6338a9bec032693f6 --- /dev/null +++ b/e28/S1-E28-I4-n16k-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:83ad0229ff8e590bbca362ce721ec5f67dfdc878cc1b73c0c6dd78ab92863a69 +size 393464 diff --git a/e28/S1-E28-I4-n16k-syn/ckpt-16000/judge_config.json b/e28/S1-E28-I4-n16k-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..b536084ff5e6b3ca067bba01e5fc3371a3ac9a53 --- /dev/null +++ b/e28/S1-E28-I4-n16k-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-I4-n16k-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-I4-n16k", + "data_manifest": { + "arm": "E28-I4-n16k", + "name": null, + "train_sha256": "85cbc9b90b873fe338831977261d4b6f6fd6a1797f359ff33ed841f40910e04b", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9805004, + "tok_per_s": 3317.237564830376, + "rows_per_s": 5.698000228794165, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5806214809417725, + "noul_T": 0.6919772888709426, + "choice_ce": 0.6720502376556396, + "choice_T": 1.0109731528245147, + "score_ce": 1.061937928199768, + "score_T": 0.9875175650796282, + "noul_auc": 0.7199829333333333, + "noul_acc": 0.707, + "noul_f1": 0.7781983345950038, + "noul_nll": 0.5961289732903242, + "noul_ece": 0.06612301194667816, + "noul_brier": 0.2043989143256383, + "noul_sharp": 0.14073404669761658, + "entropy": 0.6338372230529785 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-I4-n16k-syn/train_log.jsonl b/e28/S1-E28-I4-n16k-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..447f73fa68bcb47be066a1e0e433940e370d640f --- /dev/null +++ b/e28/S1-E28-I4-n16k-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "03:39:53", "event": "start", "data": "data/e28/E28-I4-n16k", "data_manifest": {"arm": "E28-I4-n16k", "name": null, "train_sha256": "85cbc9b90b873fe338831977261d4b6f6fd6a1797f359ff33ed841f40910e04b", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.8415350914001465} +{"time": "03:40:51", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3460.9211205447714, "loss": 0.6655305027961731, "entropy": 0.683852344751358, "sharp": 0.04635828360915184} +{"time": "03:41:47", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3477.7896486926725, "loss": 0.6887235045433044, "entropy": 0.6870831102132797, "sharp": 0.04784876946359873} +{"time": "03:42:41", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3480.6811218395515, "loss": 0.666453966870904, "entropy": 0.6452420558780432, "sharp": 0.13163881842046976} +{"time": "03:43:37", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3493.570003198514, "loss": 0.6954577565193176, "entropy": 0.6277506723999977, "sharp": 0.14208265021443367} +{"time": "03:44:33", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3498.6868602941386, "loss": 0.6740147843956947, "entropy": 0.6425900384783745, "sharp": 0.13014512229710817} +{"time": "03:45:28", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3492.5377572109182, "loss": 0.6505991294980049, "entropy": 0.6750047653913498, "sharp": 0.07610868383198977} +{"time": "03:46:24", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3488.6727468684553, "loss": 0.6631009206175804, "entropy": 0.649439662694931, "sharp": 0.12160887103527784} +{"time": "03:47:20", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3485.8073354757084, "loss": 0.6176766641438007, "entropy": 0.6248110271990299, "sharp": 0.15383952762931585} +{"time": "03:48:15", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3482.867273158658, "loss": 0.690675787627697, "entropy": 0.6781893596053123, "sharp": 0.06550095975399017} +{"time": "03:49:13", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3480.939491912783, "loss": 0.6375589147210121, "entropy": 0.6677679494023323, "sharp": 0.09793407469987869} +{"time": "03:49:33", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.6170148849487305, "noul_T": 0.6025911264945207, "noul_auc": 0.7172030851799969, "noul_acc": 0.666, "noul_f1": 0.7554904831625183, "noul_nll": 0.6154282517433166, "noul_ece": 0.11468815815448762, "noul_brier": 0.21251767169103727, "noul_sharp": 0.1113412156701088, "entropy": 0.6561915278434753} +{"time": "03:50:29", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3371.207849982969, "loss": 0.6571727991104126, "entropy": 0.6144899427890778, "sharp": 0.1682934034615755} +{"time": "03:51:25", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3379.9025055536927, "loss": 0.6583358533680439, "entropy": 0.6853883638978004, "sharp": 0.05336226616054773} +{"time": "03:52:21", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3389.2502225169687, "loss": 0.6495669819414616, "entropy": 0.6761363372206688, "sharp": 0.08090274222195148} +{"time": "03:53:16", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3398.28119595309, "loss": 0.6813245341181755, "entropy": 0.6297461539506912, "sharp": 0.13893887773156166} +{"time": "03:54:13", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3404.180958139195, "loss": 0.6616937294602394, "entropy": 0.6166309341788292, "sharp": 0.15472457837313414} +{"time": "03:55:09", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3409.472405065487, "loss": 0.6885966509580612, "entropy": 0.6724704727530479, "sharp": 0.08065671753138304} +{"time": "03:56:07", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3412.590516520836, "loss": 0.6860420759767294, "entropy": 0.6678028889000416, "sharp": 0.08608218468725681} +{"time": "03:57:04", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3416.208014186638, "loss": 0.663179449737072, "entropy": 0.6605427823960781, "sharp": 0.11132528726011515} +{"time": "03:57:59", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3420.85010985936, "loss": 0.7033318467438221, "entropy": 0.6631814539432526, "sharp": 0.09998942539095879} +{"time": "03:58:55", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3425.368342774221, "loss": 0.6226406060159206, "entropy": 0.6072829216718674, "sharp": 0.17774837650358677} +{"time": "03:59:15", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5847336053848267, "noul_T": 0.9275535058376155, "noul_auc": 0.7347555315522101, "noul_acc": 0.698, "noul_f1": 0.7756315007429421, "noul_nll": 0.5843406337350607, "noul_ece": 0.05071893966197968, "noul_brier": 0.1990491689689789, "noul_sharp": 0.18102382123470306, "entropy": 0.5964277386665344} +{"time": "04:00:12", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3370.9435271244297, "loss": 0.7263721749186516, "entropy": 0.6529713794589043, "sharp": 0.1102842865511775} +{"time": "04:01:08", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3375.557588420533, "loss": 0.6993628609925508, "entropy": 0.6746185123920441, "sharp": 0.07786909583956003} +{"time": "04:02:06", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3380.223162224938, "loss": 0.7063529342412949, "entropy": 0.6353908851742744, "sharp": 0.14549522753804922} +{"time": "04:03:03", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3383.6612284451985, "loss": 0.693148760125041, "entropy": 0.6611534617841244, "sharp": 0.10266905464231968} +{"time": "04:04:00", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3386.52920944181, "loss": 0.6568034514784813, "entropy": 0.6287561021745205, "sharp": 0.1519258301705122} +{"time": "04:04:56", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3389.5874196865316, "loss": 0.6797039657831192, "entropy": 0.6764538884162903, "sharp": 0.0698626060038805} +{"time": "04:05:54", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3392.8132348200866, "loss": 0.6612797565758228, "entropy": 0.6738704890012741, "sharp": 0.07423092145472765} +{"time": "04:06:51", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3397.7163691415735, "loss": 0.6851770021021366, "entropy": 0.6299348399043083, "sharp": 0.1401688465848565} +{"time": "04:07:49", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3399.998207927057, "loss": 0.660784050822258, "entropy": 0.6675319001078606, "sharp": 0.0878879651427269} +{"time": "04:08:44", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3404.161581688133, "loss": 0.6634425427764654, "entropy": 0.6563064642250538, "sharp": 0.11125789023935795} +{"time": "04:09:04", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5906708836555481, "noul_T": 0.7205546197607549, "noul_auc": 0.7338441545290374, "noul_acc": 0.712, "noul_f1": 0.7721518987341772, "noul_nll": 0.5916972559392453, "noul_ece": 0.06592116963863373, "noul_brier": 0.20213691848606793, "noul_sharp": 0.1481170356273651, "entropy": 0.6291436553001404} +{"time": "04:09:59", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3367.901544275212, "loss": 0.6718445867300034, "entropy": 0.6361830309033394, "sharp": 0.13705619890242815} +{"time": "04:10:55", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3372.0069523444286, "loss": 0.6805291585624218, "entropy": 0.6712068971246481, "sharp": 0.0842317882925272} +{"time": "04:11:51", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3375.925380484136, "loss": 0.6486920565366745, "entropy": 0.6755539365112782, "sharp": 0.07960772980004549} +{"time": "04:12:47", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3380.0919782022365, "loss": 0.6883015334606171, "entropy": 0.6574668884277344, "sharp": 0.10614009574055672} +{"time": "04:13:45", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3383.8031216475174, "loss": 0.6698095947504044, "entropy": 0.6590589806437492, "sharp": 0.1005021845921874} +{"time": "04:14:40", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3386.3168554291797, "loss": 0.6991751976311207, "entropy": 0.6590187549591064, "sharp": 0.1031351201236248} +{"time": "04:15:38", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3389.47322340787, "loss": 0.6418025456368923, "entropy": 0.6650706119835377, "sharp": 0.09282863326370716} +{"time": "04:16:34", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3391.3428837816277, "loss": 0.7187593802809715, "entropy": 0.6601930875331163, "sharp": 0.10888232104480267} +{"time": "04:17:32", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3393.247983711092, "loss": 0.6683227270841599, "entropy": 0.6556249186396599, "sharp": 0.11546674370765686} +{"time": "04:18:28", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3393.8678131107486, "loss": 0.6164517030119896, "entropy": 0.6564111020416021, "sharp": 0.10412486642599106} +{"time": "04:18:48", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5975287556648254, "noul_T": 0.7062614233872936, "noul_auc": 0.7328990228013028, "noul_acc": 0.714, "noul_f1": 0.7726550079491256, "noul_nll": 0.5967362135648727, "noul_ece": 0.09262246716022493, "noul_brier": 0.20415289862871958, "noul_sharp": 0.13968542218208313, "entropy": 0.6348917484283447} +{"time": "04:19:45", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3368.28908146838, "loss": 0.6283746287226677, "entropy": 0.6645361371338367, "sharp": 0.09203032217919827} +{"time": "04:20:39", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3371.493093842505, "loss": 0.6545193176716566, "entropy": 0.6466682255268097, "sharp": 0.12222914397716522} +{"time": "04:21:37", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3373.5486154377127, "loss": 0.6628632135689259, "entropy": 0.6604825593531132, "sharp": 0.10753337759524584} +{"time": "04:22:33", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3375.695666401306, "loss": 0.6971535421907902, "entropy": 0.6518079750239849, "sharp": 0.1212319266051054} +{"time": "04:23:28", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3378.37076907409, "loss": 0.5600953251123428, "entropy": 0.6287789419293404, "sharp": 0.1400765646249056} +{"time": "04:24:23", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3381.012513869204, "loss": 0.6035230588167906, "entropy": 0.6257942728698254, "sharp": 0.15616852138191462} +{"time": "04:25:19", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3382.8516055826576, "loss": 0.7030844390392303, "entropy": 0.6498725488781929, "sharp": 0.11526742111891508} +{"time": "04:26:16", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3384.686063715495, "loss": 0.6638802159577608, "entropy": 0.6555136181414127, "sharp": 0.10649732127785683} +{"time": "04:27:10", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3386.774569212602, "loss": 0.6717731803655624, "entropy": 0.6577086448669434, "sharp": 0.10903952736407518} +{"time": "04:28:07", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3388.7903357270075, "loss": 0.6797967627644539, "entropy": 0.6633354797959328, "sharp": 0.10291012562811375} +{"time": "04:28:27", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5965617299079895, "noul_T": 0.690313607653104, "noul_auc": 0.7345192486202764, "noul_acc": 0.728, "noul_f1": 0.7848101265822784, "noul_nll": 0.5964167748987674, "noul_ece": 0.09080561661720278, "noul_brier": 0.20408926837398841, "noul_sharp": 0.13725771009922028, "entropy": 0.6357023119926453} +{"time": "04:29:09", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9805004, "tok_per_s": 3317.237564830376, "rows_per_s": 5.698000228794165, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5806214809417725, "noul_T": 0.6919772888709426, "choice_ce": 0.6720502376556396, "choice_T": 1.0109731528245147, "score_ce": 1.061937928199768, "score_T": 0.9875175650796282, "noul_auc": 0.7199829333333333, "noul_acc": 0.707, "noul_f1": 0.7781983345950038, "noul_nll": 0.5961289732903242, "noul_ece": 0.06612301194667816, "noul_brier": 0.2043989143256383, "noul_sharp": 0.14073404669761658, "entropy": 0.6338372230529785}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-K49-n16k-syn/DONE b/e28/S1-E28-K49-n16k-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..42677d71395efec97c0d54ba180f176fa726fb64 --- /dev/null +++ b/e28/S1-E28-K49-n16k-syn/DONE @@ -0,0 +1 @@ +2026-10-20 20:00:07 \ No newline at end of file diff --git a/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..5860f054479f408b12c6c1a493c1a6fb6f3d88cf --- /dev/null +++ b/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "k_proj", + "out_proj", + "v_proj", + "up_proj", + "down_proj", + "o_proj", + "q_proj", + "in_proj_z", + "in_proj_qkv", + "gate_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dba2b53b3182854a8d54dfc36ee2cf7ba3e13859 --- /dev/null +++ b/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:12034f0ac78c6dad2e4932c0552f8b59c308ff82e37f56550fd098702e058e0b +size 160486456 diff --git a/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..62c1580b313c4e6f3c01f6ce87ca7fb7643c4285 --- /dev/null +++ b/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:538d7e8081c993c222fb0d374c091f57e8ff100876e3de9ddfe83dde4738b5e1 +size 176641832 diff --git a/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..af0d5bb36b09722b74a15cc85536dc858c969caa --- /dev/null +++ b/e28/S1-E28-K49-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.0002700222539715469, + -0.0002702747005969286, + -0.002342074876651168, + -0.001847754931077361, + -0.0006001496221870184, + 0.0017053456977009773, + 0.0017780398484319448, + -0.0017134061781689525, + -0.0002488352474756539, + -0.0003069194208364934, + 0.00022074082517065108, + 0.0006784375291317701, + 0.00019765846082009375, + -0.003530490444973111, + -0.003912785556167364, + -0.005074570886790752, + -0.005931418389081955, + -0.0043410505168139935, + -0.004928321111947298, + -0.004942283965647221, + -0.003849424421787262, + -0.003351946361362934, + -0.005614178255200386, + -0.003464590525254607 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-K49-n16k-syn/ckpt-16000/calibration.json b/e28/S1-E28-K49-n16k-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..35c8b2efa00eb7026d0066d0fbce042b71b47768 --- /dev/null +++ b/e28/S1-E28-K49-n16k-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.7993527359000467, + "choice": 1.0304085610572558, + "score": 0.969939319568926 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.7993527359000467, + "ce_before": 0.5712902545928955, + "ce_after": 0.5676156282424927 + }, + "choice": { + "n": 60, + "T": 1.0304085610572558, + "ce_before": 0.6637069582939148, + "ce_after": 0.6635127663612366 + }, + "score": { + "n": 53, + "T": 0.969939319568926, + "ce_before": 1.0525990724563599, + "ce_after": 1.0523039102554321 + } + }, + "source": "models/e28/S1-E28-K49-n16k-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-K49-n16k-syn/ckpt-16000/head.safetensors b/e28/S1-E28-K49-n16k-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8b3b5ab988b4d4f371511fa92daad7b976b054f9 --- /dev/null +++ b/e28/S1-E28-K49-n16k-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:19523ee572fb1acbc621816f7e29af77c55e72571d2b61d560e9e3b35792a07b +size 393464 diff --git a/e28/S1-E28-K49-n16k-syn/ckpt-16000/judge_config.json b/e28/S1-E28-K49-n16k-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..15d29f5b45bf2729d57c2be10813ff937b91ecdb --- /dev/null +++ b/e28/S1-E28-K49-n16k-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-K49-n16k-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-K49-n16k", + "data_manifest": { + "arm": "E28-K49-n16k", + "name": null, + "train_sha256": "6157990acdaffd199ebbfe903be593608c143539439dfb13877e0d854792c4e5", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9805892, + "tok_per_s": 3315.5472066789953, + "rows_per_s": 5.694580974721846, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5712902545928955, + "noul_T": 0.7993527359000467, + "choice_ce": 0.6637069582939148, + "choice_T": 1.0304085610572558, + "score_ce": 1.0525990724563599, + "score_T": 0.969939319568926, + "noul_auc": 0.7227136, + "noul_acc": 0.692, + "noul_f1": 0.7748538011695907, + "noul_nll": 0.5877951740771532, + "noul_ece": 0.03663458222150805, + "noul_brier": 0.20119157312812797, + "noul_sharp": 0.16454769670963287, + "entropy": 0.6130955815315247 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-K49-n16k-syn/train_log.jsonl b/e28/S1-E28-K49-n16k-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..f9e0e2436c4e4b539d492db76b11582a9d2c1954 --- /dev/null +++ b/e28/S1-E28-K49-n16k-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "19:10:49", "event": "start", "data": "data/e28/E28-K49-n16k", "data_manifest": {"arm": "E28-K49-n16k", "name": null, "train_sha256": "6157990acdaffd199ebbfe903be593608c143539439dfb13877e0d854792c4e5", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.3720362186431885} +{"time": "19:11:46", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3434.9213516139916, "loss": 0.664089122787118, "entropy": 0.6828516684472561, "sharp": 0.05314367637038231} +{"time": "19:12:41", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3463.9607268113164, "loss": 0.6680694594979286, "entropy": 0.6527446508407593, "sharp": 0.11137264966964722} +{"time": "19:13:36", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3468.4624590313842, "loss": 0.6776398364454508, "entropy": 0.667412905022502, "sharp": 0.08839457482099533} +{"time": "19:14:33", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3464.9374937354064, "loss": 0.6843038368970156, "entropy": 0.6782844662666321, "sharp": 0.07728857733309269} +{"time": "19:15:30", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3472.9165448564513, "loss": 0.6878192154690623, "entropy": 0.639002300798893, "sharp": 0.1427024845033884} +{"time": "19:16:27", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3477.215393576718, "loss": 0.629900187253952, "entropy": 0.6348337754607201, "sharp": 0.1422382239252329} +{"time": "19:17:23", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3476.940929194141, "loss": 0.6704541984945536, "entropy": 0.6745552457869053, "sharp": 0.07356947846710682} +{"time": "19:18:19", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3475.8854515570174, "loss": 0.6635134872049093, "entropy": 0.6785168051719666, "sharp": 0.06900553870946169} +{"time": "19:19:14", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3474.0158571452316, "loss": 0.7257608324289322, "entropy": 0.6380143240094185, "sharp": 0.14040076173841953} +{"time": "19:20:10", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3472.010203675152, "loss": 0.6541090272367001, "entropy": 0.6711159162223339, "sharp": 0.08630508556962013} +{"time": "19:20:30", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.596143901348114, "noul_T": 0.6717961099683, "noul_auc": 0.7253548463317075, "noul_acc": 0.7, "noul_f1": 0.7634069400630915, "noul_nll": 0.6004426446557045, "noul_ece": 0.07825558769702913, "noul_brier": 0.2061745583833393, "noul_sharp": 0.13537366688251495, "entropy": 0.6366795301437378} +{"time": "19:21:26", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3365.3837822178525, "loss": 0.6475662607699633, "entropy": 0.6495450772345066, "sharp": 0.12568164989352226} +{"time": "19:22:22", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3374.2350759248097, "loss": 0.6348728351294994, "entropy": 0.6661745645105839, "sharp": 0.09748838935047388} +{"time": "19:23:17", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3382.3398277919205, "loss": 0.6498086452484131, "entropy": 0.63322713971138, "sharp": 0.1472855107858777} +{"time": "19:24:15", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3390.8303031200835, "loss": 0.6808188632130623, "entropy": 0.6453310810029507, "sharp": 0.1268774587661028} +{"time": "19:25:10", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3394.710552235921, "loss": 0.695251002907753, "entropy": 0.6800846755504608, "sharp": 0.06357597652822733} +{"time": "19:26:08", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3396.398923389304, "loss": 0.7511026412248611, "entropy": 0.6737593710422516, "sharp": 0.07718713395297527} +{"time": "19:27:06", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3404.8515847933345, "loss": 0.6510995328426361, "entropy": 0.6576029844582081, "sharp": 0.11304814368486404} +{"time": "19:28:03", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3411.5949081538024, "loss": 0.6177741885185242, "entropy": 0.6674432493746281, "sharp": 0.09408258460462093} +{"time": "19:28:59", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3415.172289577324, "loss": 0.6581366881728172, "entropy": 0.6552661098539829, "sharp": 0.11224623117595911} +{"time": "19:29:55", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3417.583324818163, "loss": 0.648795697838068, "entropy": 0.6574468165636063, "sharp": 0.11001930385828018} +{"time": "19:30:15", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.597247302532196, "noul_T": 0.8928264681414159, "noul_auc": 0.7392111525543872, "noul_acc": 0.694, "noul_f1": 0.7454242928452579, "noul_nll": 0.5918655084520579, "noul_ece": 0.07405886006355285, "noul_brier": 0.20254889313770436, "noul_sharp": 0.16845528781414032, "entropy": 0.6127938032150269} +{"time": "19:31:12", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3366.325832260762, "loss": 0.6964229941368103, "entropy": 0.6655927114188671, "sharp": 0.0993328457698226} +{"time": "19:32:08", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3370.6176630463506, "loss": 0.6881728582084179, "entropy": 0.6645393818616867, "sharp": 0.10558377578854561} +{"time": "19:33:03", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3374.7572577462092, "loss": 0.6604227870702744, "entropy": 0.6673952043056488, "sharp": 0.09692526888102293} +{"time": "19:33:59", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3378.805740586169, "loss": 0.6583612523972988, "entropy": 0.6307640299201012, "sharp": 0.14764054957777262} +{"time": "19:34:55", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3382.971407002731, "loss": 0.6409013457596302, "entropy": 0.6622830331325531, "sharp": 0.10489773005247116} +{"time": "19:35:51", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3386.6603332405866, "loss": 0.6293181031942368, "entropy": 0.6704294756054878, "sharp": 0.08355523180216551} +{"time": "19:36:49", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3391.065692258973, "loss": 0.6378923691809177, "entropy": 0.6502885669469833, "sharp": 0.11384815070778131} +{"time": "19:37:45", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3394.0913886748062, "loss": 0.6972200945019722, "entropy": 0.6337089687585831, "sharp": 0.14423477463424206} +{"time": "19:38:40", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3396.782599530277, "loss": 0.6903045773506165, "entropy": 0.6570149958133698, "sharp": 0.11309993173927069} +{"time": "19:39:35", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3399.0757884792406, "loss": 0.6717061102390289, "entropy": 0.6500616893172264, "sharp": 0.11313075758516788} +{"time": "19:39:55", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5855368971824646, "noul_T": 0.829065649674393, "noul_auc": 0.7436498962042835, "noul_acc": 0.704, "noul_f1": 0.7597402597402597, "noul_nll": 0.5846324564218521, "noul_ece": 0.0646208912730217, "noul_brier": 0.19963929823198628, "noul_sharp": 0.16202566027641296, "entropy": 0.6102099418640137} +{"time": "19:40:52", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3362.8290115959435, "loss": 0.6498517841100693, "entropy": 0.6633243784308434, "sharp": 0.10541055910289288} +{"time": "19:41:49", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3367.0052792622764, "loss": 0.6335206367075443, "entropy": 0.635753620415926, "sharp": 0.1384343123063445} +{"time": "19:42:43", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3371.117695434039, "loss": 0.6672131400555372, "entropy": 0.6395172253251076, "sharp": 0.13990065176039934} +{"time": "19:43:40", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3375.032179190977, "loss": 0.690017968416214, "entropy": 0.6612017955631018, "sharp": 0.09899160452187061} +{"time": "19:44:35", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3377.799739720152, "loss": 0.6291274204850197, "entropy": 0.6716634333133698, "sharp": 0.08885768707841635} +{"time": "19:45:33", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3381.109267671128, "loss": 0.655661053955555, "entropy": 0.6586408466100693, "sharp": 0.11143565457314253} +{"time": "19:46:29", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3383.708805706462, "loss": 0.6190427467226982, "entropy": 0.6555690839886665, "sharp": 0.12048686016350985} +{"time": "19:47:25", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3386.339727929053, "loss": 0.6927382573485374, "entropy": 0.6711444482207298, "sharp": 0.08978140167891979} +{"time": "19:48:22", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3389.532834932975, "loss": 0.7094521746039391, "entropy": 0.6711103022098541, "sharp": 0.08440647460520267} +{"time": "19:49:19", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3391.7094904308433, "loss": 0.6144724674522877, "entropy": 0.6697610095143318, "sharp": 0.09369289875030518} +{"time": "19:49:38", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.603701651096344, "noul_T": 0.5747209466740207, "noul_auc": 0.7406457274982701, "noul_acc": 0.708, "noul_f1": 0.7697160883280757, "noul_nll": 0.6023463318645954, "noul_ece": 0.09137966835498812, "noul_brier": 0.20650892629893494, "noul_sharp": 0.1186714619398117, "entropy": 0.6513742208480835} +{"time": "19:50:34", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3365.7284379551515, "loss": 0.6685272082686424, "entropy": 0.6564998105168343, "sharp": 0.12047909386456013} +{"time": "19:51:29", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3368.71444688531, "loss": 0.6682560443878174, "entropy": 0.6719005294144154, "sharp": 0.08909965679049492} +{"time": "19:52:26", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3371.298870902973, "loss": 0.6508132740855217, "entropy": 0.6494929566979408, "sharp": 0.12502507399767637} +{"time": "19:53:25", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3374.331851169251, "loss": 0.6514615565538406, "entropy": 0.6565827056765556, "sharp": 0.11088694352656603} +{"time": "19:54:20", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3376.5008622645064, "loss": 0.6437789127230644, "entropy": 0.6473867297172546, "sharp": 0.12707093358039856} +{"time": "19:55:17", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3379.120564823442, "loss": 0.6619072780013084, "entropy": 0.6485094390809536, "sharp": 0.12322958279401064} +{"time": "19:56:14", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3381.248979007961, "loss": 0.732081837952137, "entropy": 0.6401232481002808, "sharp": 0.1366329463198781} +{"time": "19:57:11", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3382.966567943013, "loss": 0.652949184179306, "entropy": 0.6496239714324474, "sharp": 0.12557489704340696} +{"time": "19:58:07", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3384.4408259907223, "loss": 0.6448281034827232, "entropy": 0.6495150253176689, "sharp": 0.11746042128652334} +{"time": "19:59:04", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3387.002525756481, "loss": 0.671174150891602, "entropy": 0.6570743750780821, "sharp": 0.11146005801856518} +{"time": "19:59:24", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5843033790588379, "noul_T": 0.7713429009339572, "noul_auc": 0.7400718975207168, "noul_acc": 0.7, "noul_f1": 0.7706422018348624, "noul_nll": 0.5848417781591415, "noul_ece": 0.04870740342140199, "noul_brier": 0.1993646111069366, "noul_sharp": 0.1581285446882248, "entropy": 0.6175839304924011} +{"time": "20:00:07", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9805892, "tok_per_s": 3315.5472066789953, "rows_per_s": 5.694580974721846, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5712902545928955, "noul_T": 0.7993527359000467, "choice_ce": 0.6637069582939148, "choice_T": 1.0304085610572558, "score_ce": 1.0525990724563599, "score_T": 0.969939319568926, "noul_auc": 0.7227136, "noul_acc": 0.692, "noul_f1": 0.7748538011695907, "noul_nll": 0.5877951740771532, "noul_ece": 0.03663458222150805, "noul_brier": 0.20119157312812797, "noul_sharp": 0.16454769670963287, "entropy": 0.6130955815315247}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-K784-n16k-syn/DONE b/e28/S1-E28-K784-n16k-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..0b2b668bbb610866c142420b590a31827423e65f --- /dev/null +++ b/e28/S1-E28-K784-n16k-syn/DONE @@ -0,0 +1 @@ +2026-10-20 21:13:49 \ No newline at end of file diff --git a/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..85a6ed1571b96fcb10f02d90356b0e75716ecd39 --- /dev/null +++ b/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "k_proj", + "o_proj", + "q_proj", + "up_proj", + "out_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "down_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7b57733163ecc53ff0e14be7a6d35c8c870555ab --- /dev/null +++ b/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:173593e7ddad461d52171d9c3cec6f1a0c3f18243fb8f811da7971aa897a17e2 +size 160486456 diff --git a/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..455a829181de16b62d330a966cfdd773a9c132b3 --- /dev/null +++ b/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c678be25073fcef510677f7deb15a4119b3b16dd8048091100adaeb6f0249c7b +size 176641832 diff --git a/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..47ee271c1f902beea276d7a707513815671d258f --- /dev/null +++ b/e28/S1-E28-K784-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.0004172554472461343, + -0.0004175095818936825, + -0.0028529732953757048, + -0.0017298638122156262, + -0.0007133790641091764, + 0.002251736354082823, + 0.0014804585371166468, + -0.0024324371479451656, + -0.0003944132477045059, + -0.00041116756619885564, + 7.382812327705324e-05, + 0.0001892250293167308, + -0.00026986145530827343, + -0.0028447022195905447, + -0.004384391009807587, + -0.00518037099391222, + -0.00601166719570756, + -0.00438054371625185, + -0.004936442244797945, + -0.004975837655365467, + -0.0037860500160604715, + -0.0033871843479573727, + -0.005675188731402159, + -0.0034968850668519735 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-K784-n16k-syn/ckpt-16000/calibration.json b/e28/S1-E28-K784-n16k-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..bc262e41ec1aecfb1541f405f0726d8e3679fbc9 --- /dev/null +++ b/e28/S1-E28-K784-n16k-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.8418124600419037, + "choice": 0.9804189486029293, + "score": 0.9814221537970804 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.8418124600419037, + "ce_before": 0.5706677436828613, + "ce_after": 0.5683627724647522 + }, + "choice": { + "n": 60, + "T": 0.9804189486029293, + "ce_before": 0.6730120182037354, + "ce_after": 0.672924816608429 + }, + "score": { + "n": 53, + "T": 0.9814221537970804, + "ce_before": 1.0602589845657349, + "ce_after": 1.0601489543914795 + } + }, + "source": "models/e28/S1-E28-K784-n16k-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-K784-n16k-syn/ckpt-16000/head.safetensors b/e28/S1-E28-K784-n16k-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c18fdfc372ec4ef0a697f40d395c4a843301e20e --- /dev/null +++ b/e28/S1-E28-K784-n16k-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a433e07a182bb8e9f1c4d8ecbbe20db574c90d99838401a71fc7963f57cf463 +size 393464 diff --git a/e28/S1-E28-K784-n16k-syn/ckpt-16000/judge_config.json b/e28/S1-E28-K784-n16k-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..832e03ebdc540c957c53ad703cfc4c2b43e4141b --- /dev/null +++ b/e28/S1-E28-K784-n16k-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-K784-n16k-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-K784-n16k", + "data_manifest": { + "arm": "E28-K784-n16k", + "name": null, + "train_sha256": "58ea47d9812fd2997f8fba8d871afb978eac3cdbc72f2323dc34ffe32cef9e9c", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9813269, + "tok_per_s": 3321.888855565745, + "rows_per_s": 5.7011839865335645, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5706677436828613, + "noul_T": 0.8418124600419037, + "choice_ce": 0.6730120182037354, + "choice_T": 0.9804189486029293, + "score_ce": 1.0602589845657349, + "score_T": 0.9814221537970804, + "noul_auc": 0.7241045333333334, + "noul_acc": 0.69, + "noul_f1": 0.7693452380952381, + "noul_nll": 0.5871793775931, + "noul_ece": 0.055497723907232295, + "noul_brier": 0.2007807299090352, + "noul_sharp": 0.1736905723810196, + "entropy": 0.6075435876846313 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-K784-n16k-syn/train_log.jsonl b/e28/S1-E28-K784-n16k-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..db7209dac2f35781a57bf0ca9f850b7172b88e39 --- /dev/null +++ b/e28/S1-E28-K784-n16k-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "20:24:35", "event": "start", "data": "data/e28/E28-K784-n16k", "data_manifest": {"arm": "E28-K784-n16k", "name": null, "train_sha256": "58ea47d9812fd2997f8fba8d871afb978eac3cdbc72f2323dc34ffe32cef9e9c", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.359998941421509} +{"time": "20:25:32", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3424.5696254623363, "loss": 0.6944657899439335, "entropy": 0.6753967553377151, "sharp": 0.07545473799109459} +{"time": "20:26:28", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3437.405980326958, "loss": 0.688731461763382, "entropy": 0.6605916619300842, "sharp": 0.10509055014699697} +{"time": "20:27:25", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3455.062864489569, "loss": 0.7530146073549986, "entropy": 0.6782102342694998, "sharp": 0.06741233170032501} +{"time": "20:28:21", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3479.4317003208344, "loss": 0.7324719158932567, "entropy": 0.6049990076571703, "sharp": 0.16194368246942759} +{"time": "20:29:15", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3480.94506936666, "loss": 0.6750261709094048, "entropy": 0.6649212390184402, "sharp": 0.09504927042871714} +{"time": "20:30:10", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3483.0041905666926, "loss": 0.6219950616359711, "entropy": 0.6411548126488924, "sharp": 0.1451119352132082} +{"time": "20:31:06", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3481.6306597222406, "loss": 0.6486634016036987, "entropy": 0.6432040445506573, "sharp": 0.1348405471071601} +{"time": "20:32:01", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3482.142697058199, "loss": 0.6707365363836288, "entropy": 0.6626019589602947, "sharp": 0.1020119171589613} +{"time": "20:32:58", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3477.7902652133653, "loss": 0.7282356321811676, "entropy": 0.6109040975570679, "sharp": 0.15927963424474} +{"time": "20:33:54", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3478.5140138513893, "loss": 0.5974017158150673, "entropy": 0.6565393358469009, "sharp": 0.09568904340267181} +{"time": "20:34:14", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.6034408211708069, "noul_T": 0.6544642730117535, "noul_auc": 0.7351099559501105, "noul_acc": 0.682, "noul_f1": 0.7318718381112985, "noul_nll": 0.6036328387260437, "noul_ece": 0.09518015837669373, "noul_brier": 0.20739030128076588, "noul_sharp": 0.12559916079044342, "entropy": 0.6433941125869751} +{"time": "20:35:11", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3375.3782846685394, "loss": 0.6595800817012787, "entropy": 0.6494352594017982, "sharp": 0.1236268887296319} +{"time": "20:36:05", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3380.870364819149, "loss": 0.6773656010627747, "entropy": 0.6566512435674667, "sharp": 0.10709482338279486} +{"time": "20:37:01", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3388.197716255851, "loss": 0.6480531673878431, "entropy": 0.6688263937830925, "sharp": 0.0971175292506814} +{"time": "20:37:56", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3395.0729012372844, "loss": 0.6688768938183784, "entropy": 0.6782380566000938, "sharp": 0.06073042098432779} +{"time": "20:38:51", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3400.823949972459, "loss": 0.6649162173271179, "entropy": 0.6737324483692646, "sharp": 0.08419746160507202} +{"time": "20:39:47", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3405.395449448614, "loss": 0.684875650331378, "entropy": 0.5864386931061745, "sharp": 0.20481269899755716} +{"time": "20:40:43", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3410.1181593538017, "loss": 0.6329841613769531, "entropy": 0.6537240073084831, "sharp": 0.1290738582611084} +{"time": "20:41:40", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3416.007569980218, "loss": 0.6524203196167946, "entropy": 0.6785571649670601, "sharp": 0.06844785157591105} +{"time": "20:42:36", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3420.373979144155, "loss": 0.6325470842421055, "entropy": 0.6345253624022007, "sharp": 0.1422006618231535} +{"time": "20:43:32", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3424.6848498831855, "loss": 0.6920699700713158, "entropy": 0.6816733926534653, "sharp": 0.057585119269788265} +{"time": "20:43:52", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.6119363903999329, "noul_T": 0.5168469245934453, "noul_auc": 0.7344686165634334, "noul_acc": 0.692, "noul_f1": 0.7516129032258064, "noul_nll": 0.611948264837265, "noul_ece": 0.10910244286060335, "noul_brier": 0.21063229489141816, "noul_sharp": 0.10665083676576614, "entropy": 0.6603275537490845} +{"time": "20:44:48", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3372.867926344917, "loss": 0.7385538145899773, "entropy": 0.6573869287967682, "sharp": 0.11542150098830462} +{"time": "20:45:44", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3376.0889567300464, "loss": 0.6689097341150045, "entropy": 0.6126233711838722, "sharp": 0.16439730487763882} +{"time": "20:46:40", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3380.416356964498, "loss": 0.6923186182975769, "entropy": 0.6697365418076515, "sharp": 0.087502621114254} +{"time": "20:47:37", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3385.189808595111, "loss": 0.6744275726377964, "entropy": 0.6651755645871162, "sharp": 0.10035560559481382} +{"time": "20:48:34", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3389.6577954577724, "loss": 0.6883103996515274, "entropy": 0.658838551491499, "sharp": 0.10407024342566729} +{"time": "20:49:30", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3394.676381916464, "loss": 0.6213240921497345, "entropy": 0.6496841311454773, "sharp": 0.11821815371513367} +{"time": "20:50:27", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3398.397590041819, "loss": 0.6659102812409401, "entropy": 0.666113592684269, "sharp": 0.0963958278298378} +{"time": "20:51:24", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3400.046962502907, "loss": 0.6721932664513588, "entropy": 0.6547729298472404, "sharp": 0.10343061480671167} +{"time": "20:52:20", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3402.100118395323, "loss": 0.678150936961174, "entropy": 0.6378898657858372, "sharp": 0.14761252515017986} +{"time": "20:53:18", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3403.940140609903, "loss": 0.7239544056355953, "entropy": 0.6690349001437426, "sharp": 0.09050349146127701} +{"time": "20:53:38", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5897800326347351, "noul_T": 0.7661378062804992, "noul_auc": 0.7363757573711835, "noul_acc": 0.704, "noul_f1": 0.782991202346041, "noul_nll": 0.5901746745109558, "noul_ece": 0.06455341768264773, "noul_brier": 0.20134665502048346, "noul_sharp": 0.1566057950258255, "entropy": 0.6231749653816223} +{"time": "20:54:33", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3369.5025040409146, "loss": 0.6462365612387657, "entropy": 0.6799611151218414, "sharp": 0.06940750032663345} +{"time": "20:55:30", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3373.040967885151, "loss": 0.6322518503293395, "entropy": 0.6327041499316692, "sharp": 0.13781048823148012} +{"time": "20:56:26", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3376.701122121126, "loss": 0.670484684407711, "entropy": 0.6399454101920128, "sharp": 0.1381393214687705} +{"time": "20:57:23", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3378.7883179091546, "loss": 0.6279081720858812, "entropy": 0.6520563047379255, "sharp": 0.12240668386220932} +{"time": "20:58:19", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3382.5649057984197, "loss": 0.6337915984913707, "entropy": 0.6473472528159618, "sharp": 0.11644069105386734} +{"time": "20:59:15", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3385.3600322923685, "loss": 0.6856397464871407, "entropy": 0.6501713395118713, "sharp": 0.12376044876873493} +{"time": "21:00:10", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3388.536919103251, "loss": 0.6835046634078026, "entropy": 0.655614759773016, "sharp": 0.11434797756373882} +{"time": "21:01:07", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3391.1356602206815, "loss": 0.6484493538737297, "entropy": 0.6543717533349991, "sharp": 0.11437278520315886} +{"time": "21:02:02", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3394.3589473648353, "loss": 0.6754177175462246, "entropy": 0.658662423491478, "sharp": 0.10447821952402592} +{"time": "21:02:57", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3396.085504382826, "loss": 0.6325298137962818, "entropy": 0.6524992398917675, "sharp": 0.11058189440518618} +{"time": "21:03:17", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5929350852966309, "noul_T": 0.6461752925714951, "noul_auc": 0.7384179169971814, "noul_acc": 0.712, "noul_f1": 0.7811550151975684, "noul_nll": 0.5933116057515144, "noul_ece": 0.07798224031925202, "noul_brier": 0.2025899211516692, "noul_sharp": 0.13630816340446472, "entropy": 0.6387923359870911} +{"time": "21:04:13", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3369.8124019845613, "loss": 0.6976286098361015, "entropy": 0.6662681847810745, "sharp": 0.09175940789282322} +{"time": "21:05:11", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3372.0147615754227, "loss": 0.6334326677024364, "entropy": 0.642073929309845, "sharp": 0.13747270964086056} +{"time": "21:06:09", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3374.7580375309017, "loss": 0.7096732929348946, "entropy": 0.6245371848344803, "sharp": 0.14818480424582958} +{"time": "21:07:06", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3378.0885516737326, "loss": 0.6720178723335266, "entropy": 0.6564519181847572, "sharp": 0.11181358527392149} +{"time": "21:08:02", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3381.258339194868, "loss": 0.6509527340531349, "entropy": 0.6655087843537331, "sharp": 0.10003454051911831} +{"time": "21:09:01", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3384.529396250622, "loss": 0.6573914773762226, "entropy": 0.6621555741876364, "sharp": 0.09534230921417475} +{"time": "21:09:56", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3386.2328251740246, "loss": 0.6734531074762344, "entropy": 0.643253929913044, "sharp": 0.1385921509936452} +{"time": "21:10:53", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3388.038501208452, "loss": 0.6494116829708219, "entropy": 0.6438987590372562, "sharp": 0.13363000564277172} +{"time": "21:11:51", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3391.0623287008616, "loss": 0.6371779330074787, "entropy": 0.6621463187038898, "sharp": 0.10465321317315102} +{"time": "21:12:46", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3393.5470242539227, "loss": 0.6405036598443985, "entropy": 0.642711952328682, "sharp": 0.1329815313220024} +{"time": "21:13:06", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5844843983650208, "noul_T": 0.8298552939379343, "noul_auc": 0.7383672849403387, "noul_acc": 0.702, "noul_f1": 0.7697063369397218, "noul_nll": 0.58454058419168, "noul_ece": 0.0586618243455887, "noul_brier": 0.19900052296520568, "noul_sharp": 0.16949325799942017, "entropy": 0.6107252240180969} +{"time": "21:13:49", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9813269, "tok_per_s": 3321.888855565745, "rows_per_s": 5.7011839865335645, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5706677436828613, "noul_T": 0.8418124600419037, "choice_ce": 0.6730120182037354, "choice_T": 0.9804189486029293, "score_ce": 1.0602589845657349, "score_T": 0.9814221537970804, "noul_auc": 0.7241045333333334, "noul_acc": 0.69, "noul_f1": 0.7693452380952381, "noul_nll": 0.5871793775931, "noul_ece": 0.055497723907232295, "noul_brier": 0.2007807299090352, "noul_sharp": 0.1736905723810196, "entropy": 0.6075435876846313}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-M1-n16k-syn/DONE b/e28/S1-E28-M1-n16k-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..21f0680f1637be14887ae396e542c58160df82c0 --- /dev/null +++ b/e28/S1-E28-M1-n16k-syn/DONE @@ -0,0 +1 @@ +2026-10-20 07:53:21 \ No newline at end of file diff --git a/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..84ce911bb9c9cbd22ac35ace75a2bff9ebebee30 --- /dev/null +++ b/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "o_proj", + "down_proj", + "k_proj", + "out_proj", + "v_proj", + "gate_proj", + "up_proj", + "in_proj_z", + "in_proj_qkv", + "q_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dc6fc622078d6cde925cf47a5f0d81e13c6e6d1a --- /dev/null +++ b/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:18a03756c8b87cdd11887bb551d56d27f9b1b17325cb4cce8255fa4a38fd6cf3 +size 160486456 diff --git a/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..64ea89a8167e9befd7197d737de48aa9ea842bcb --- /dev/null +++ b/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bbb1d1238aab2fd7759fd90e17b2bf029c4a56279a948ab03f7b22c00f92ed28 +size 176641832 diff --git a/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..3dd34af54e0f0be1ae66d5ce0189c7433fc1335b --- /dev/null +++ b/e28/S1-E28-M1-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.00038641307037323713, + -0.00038666525506414473, + -0.002495148917660117, + -0.0014474818017333746, + -0.000429368345066905, + 0.0021123234182596207, + 0.001308112172409892, + -0.0029148005414754152, + -0.0007643187418580055, + -0.0004906764952465892, + -9.819795377552509e-05, + -3.398029366508126e-05, + -0.0010864520445466042, + -0.0033147563226521015, + -0.004465482663363218, + -0.005089274607598782, + -0.005941230803728104, + -0.00434478186070919, + -0.00493140984326601, + -0.004969695582985878, + -0.0038919798098504543, + -0.0033669909462332726, + -0.005611908622086048, + -0.0034778607077896595 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-M1-n16k-syn/ckpt-16000/calibration.json b/e28/S1-E28-M1-n16k-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..df349bf1f6b9b5fb09a339b922a4a891a0f8908c --- /dev/null +++ b/e28/S1-E28-M1-n16k-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.7367525983104597, + "choice": 1.0616362628517306, + "score": 0.9881188547581379 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.7367525983104597, + "ce_before": 0.5789893269538879, + "ce_after": 0.5727285742759705 + }, + "choice": { + "n": 60, + "T": 1.0616362628517306, + "ce_before": 0.6876552104949951, + "ce_after": 0.686862051486969 + }, + "score": { + "n": 53, + "T": 0.9881188547581379, + "ce_before": 1.0693273544311523, + "ce_after": 1.0692837238311768 + } + }, + "source": "models/e28/S1-E28-M1-n16k-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-M1-n16k-syn/ckpt-16000/head.safetensors b/e28/S1-E28-M1-n16k-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..09707dbb7462326c759444cfe5ed098cff30f49a --- /dev/null +++ b/e28/S1-E28-M1-n16k-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b8a4b9d819efc8cb7567616c7be0e11a425b059677093e4f5de55b98ee2514e +size 393464 diff --git a/e28/S1-E28-M1-n16k-syn/ckpt-16000/judge_config.json b/e28/S1-E28-M1-n16k-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..e5517dc73d9ba8e4d46bfb150ae9079cb160066b --- /dev/null +++ b/e28/S1-E28-M1-n16k-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-M1-n16k-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-M1-n16k", + "data_manifest": { + "arm": "E28-M1-n16k", + "name": null, + "train_sha256": "effcb09e4490f8a65399b09ff8231663e37c06336bdf6711b7c924d5a3d7e1fd", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9812951, + "tok_per_s": 3324.467535176867, + "rows_per_s": 5.70579453625253, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5789893269538879, + "noul_T": 0.7367525983104597, + "choice_ce": 0.6876552104949951, + "choice_T": 1.0616362628517306, + "score_ce": 1.0693273544311523, + "score_T": 0.9881188547581379, + "noul_auc": 0.7117098666666666, + "noul_acc": 0.683, + "noul_f1": 0.7643122676579925, + "noul_nll": 0.5978014750629663, + "noul_ece": 0.05898461356759074, + "noul_brier": 0.2051665544090499, + "noul_sharp": 0.1491762399673462, + "entropy": 0.6274650692939758 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-M1-n16k-syn/train_log.jsonl b/e28/S1-E28-M1-n16k-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..6aec064259d91d3decb551312fda579b84f754a2 --- /dev/null +++ b/e28/S1-E28-M1-n16k-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "07:04:09", "event": "start", "data": "data/e28/E28-M1-n16k", "data_manifest": {"arm": "E28-M1-n16k", "name": null, "train_sha256": "effcb09e4490f8a65399b09ff8231663e37c06336bdf6711b7c924d5a3d7e1fd", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.7328362464904785} +{"time": "07:05:05", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3446.0968881672393, "loss": 0.6746844127774239, "entropy": 0.6731318980455399, "sharp": 0.08309441059827805} +{"time": "07:06:01", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3455.966878869738, "loss": 0.6705286614596844, "entropy": 0.6742783151566982, "sharp": 0.07938473857939243} +{"time": "07:06:58", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3467.242066110609, "loss": 0.6477400995790958, "entropy": 0.6582728773355484, "sharp": 0.12351613491773605} +{"time": "07:07:55", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3474.1409740425256, "loss": 0.6730263754725456, "entropy": 0.6673815324902534, "sharp": 0.08652760181576014} +{"time": "07:08:51", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3482.011225265272, "loss": 0.616624265909195, "entropy": 0.6404372155666351, "sharp": 0.13966528605669737} +{"time": "07:09:47", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3484.736105443631, "loss": 0.6550705879926682, "entropy": 0.6744649335741997, "sharp": 0.06973834708333015} +{"time": "07:10:42", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3485.497682176669, "loss": 0.6535475496202707, "entropy": 0.6562256775796413, "sharp": 0.11771588027477264} +{"time": "07:11:37", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3488.1530254401728, "loss": 0.6792320758104324, "entropy": 0.6711824610829353, "sharp": 0.08959460631012917} +{"time": "07:12:33", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3488.0733260755133, "loss": 0.7091983314603567, "entropy": 0.6601617261767387, "sharp": 0.1042260155081749} +{"time": "07:13:26", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3487.838987523046, "loss": 0.6305186077952385, "entropy": 0.684823676943779, "sharp": 0.054437268525362015} +{"time": "07:13:46", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.6269927620887756, "noul_T": 0.5366912774632318, "noul_auc": 0.6926465376111797, "noul_acc": 0.64, "noul_f1": 0.7560975609756098, "noul_nll": 0.6309350549578666, "noul_ece": 0.10766886830329897, "noul_brier": 0.21984749257427136, "noul_sharp": 0.09761784970760345, "entropy": 0.6662240624427795} +{"time": "07:14:43", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3375.3770441728407, "loss": 0.6715235412120819, "entropy": 0.6623129025101662, "sharp": 0.0957430973649025} +{"time": "07:15:40", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3382.6136560024543, "loss": 0.6701445579528809, "entropy": 0.667598083615303, "sharp": 0.09375184308737516} +{"time": "07:16:34", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3390.2975928204337, "loss": 0.6560290679335594, "entropy": 0.6815897077322006, "sharp": 0.058488961309194565} +{"time": "07:17:29", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3396.818030841984, "loss": 0.6630296781659126, "entropy": 0.6631457731127739, "sharp": 0.09869148395955563} +{"time": "07:18:24", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3403.8305711393, "loss": 0.6338046081364155, "entropy": 0.6597563959658146, "sharp": 0.10209724213927984} +{"time": "07:19:19", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3407.174850244502, "loss": 0.729123055934906, "entropy": 0.659957654774189, "sharp": 0.11021491046994925} +{"time": "07:20:15", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3411.4734818544052, "loss": 0.6577142588794231, "entropy": 0.6585182845592499, "sharp": 0.10815158300101757} +{"time": "07:21:09", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3416.4781169599205, "loss": 0.6326540559530258, "entropy": 0.6541057303547859, "sharp": 0.11720884498208761} +{"time": "07:22:05", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3420.1016819644924, "loss": 0.6338678225874901, "entropy": 0.6491868942975998, "sharp": 0.1182715306058526} +{"time": "07:23:01", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3423.2723758696843, "loss": 0.661566399037838, "entropy": 0.6393864266574383, "sharp": 0.13749093748629093} +{"time": "07:23:21", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.6218163967132568, "noul_T": 0.7250602532308724, "noul_auc": 0.7227219793758757, "noul_acc": 0.668, "noul_f1": 0.7077464788732394, "noul_nll": 0.6213540439605713, "noul_ece": 0.10823047399520873, "noul_brier": 0.21546033920354574, "noul_sharp": 0.12530694901943207, "entropy": 0.6498457193374634} +{"time": "07:24:17", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3368.729382545382, "loss": 0.6984150521457195, "entropy": 0.6415674220770597, "sharp": 0.12233787216246128} +{"time": "07:25:12", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3372.0364200778063, "loss": 0.6536335498094559, "entropy": 0.6383944228291512, "sharp": 0.1306010838598013} +{"time": "07:26:09", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3376.4201098246963, "loss": 0.6883286470547318, "entropy": 0.6340217981487513, "sharp": 0.14418576750904322} +{"time": "07:27:06", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3381.409190965922, "loss": 0.6877023801207542, "entropy": 0.675931703299284, "sharp": 0.0656200721859932} +{"time": "07:28:02", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3384.3498174851056, "loss": 0.6266786996275187, "entropy": 0.6469290722161531, "sharp": 0.10849734861403704} +{"time": "07:28:58", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3387.873905537094, "loss": 0.6665300019085407, "entropy": 0.6216453909873962, "sharp": 0.16135514993220568} +{"time": "07:29:54", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3391.844349588241, "loss": 0.6446754634380341, "entropy": 0.6590540260076523, "sharp": 0.10455439705401659} +{"time": "07:30:51", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3395.371261907808, "loss": 0.6729646781459451, "entropy": 0.6525364629924297, "sharp": 0.10793717578053474} +{"time": "07:31:48", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3399.397180123904, "loss": 0.6489990577101707, "entropy": 0.6228718124330044, "sharp": 0.15089535992592573} +{"time": "07:32:44", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3402.6750675224453, "loss": 0.6789086982607841, "entropy": 0.66966437920928, "sharp": 0.08231113944202662} +{"time": "07:33:04", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5977752208709717, "noul_T": 0.6967208066675342, "noul_auc": 0.7206460650453158, "noul_acc": 0.694, "noul_f1": 0.7627906976744186, "noul_nll": 0.6016546871960163, "noul_ece": 0.07545246279239655, "noul_brier": 0.20635738729525763, "noul_sharp": 0.13978390395641327, "entropy": 0.6364590525627136} +{"time": "07:34:00", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3368.3661603987175, "loss": 0.6646943017840385, "entropy": 0.6535074710845947, "sharp": 0.10861485917121172} +{"time": "07:34:57", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3373.434844131668, "loss": 0.6504230871796608, "entropy": 0.6552077606320381, "sharp": 0.10740214306861162} +{"time": "07:35:53", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3378.183418232348, "loss": 0.649666104465723, "entropy": 0.6760581973940134, "sharp": 0.07033148128539324} +{"time": "07:36:49", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3381.7250963143483, "loss": 0.6998333558440208, "entropy": 0.6571719124913216, "sharp": 0.11148007400333881} +{"time": "07:37:46", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3384.7739200985934, "loss": 0.6276538409292698, "entropy": 0.6426404789090157, "sharp": 0.13806099072098732} +{"time": "07:38:44", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3389.268190639088, "loss": 0.676094863563776, "entropy": 0.6523648872971535, "sharp": 0.111043406650424} +{"time": "07:39:40", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3392.6640707864462, "loss": 0.6718256697058678, "entropy": 0.6374876908957958, "sharp": 0.12405864149332047} +{"time": "07:40:36", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3395.7082119169695, "loss": 0.683204784989357, "entropy": 0.6653823219239712, "sharp": 0.08760177437216043} +{"time": "07:41:33", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3398.4745586671, "loss": 0.7133071050047874, "entropy": 0.6577103659510612, "sharp": 0.1051099132746458} +{"time": "07:42:29", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3400.0622879602174, "loss": 0.6433428041636944, "entropy": 0.6609953939914703, "sharp": 0.10414201952517033} +{"time": "07:42:49", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5938864946365356, "noul_T": 0.651951104785905, "noul_auc": 0.7269244400938382, "noul_acc": 0.708, "noul_f1": 0.7760736196319018, "noul_nll": 0.5989234205782413, "noul_ece": 0.07780623638629912, "noul_brier": 0.20521584604575527, "noul_sharp": 0.1347310096025467, "entropy": 0.6380991339683533} +{"time": "07:43:45", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3373.608537212984, "loss": 0.6561453379690647, "entropy": 0.6737835109233856, "sharp": 0.08076497726142406} +{"time": "07:44:42", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3375.8502201059164, "loss": 0.6280458942055702, "entropy": 0.6441082656383514, "sharp": 0.1295849271118641} +{"time": "07:45:36", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3378.672046250685, "loss": 0.6869309283792973, "entropy": 0.6584954634308815, "sharp": 0.10513441078364849} +{"time": "07:46:36", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3382.5868029444664, "loss": 0.705086924135685, "entropy": 0.655186839401722, "sharp": 0.10842266213148832} +{"time": "07:47:32", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3384.6666691055857, "loss": 0.6374431382864714, "entropy": 0.647118017077446, "sharp": 0.11876463051885366} +{"time": "07:48:31", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3387.4439816996373, "loss": 0.5949771553277969, "entropy": 0.6522279791533947, "sharp": 0.11571999453008175} +{"time": "07:49:27", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3390.2083176148903, "loss": 0.6572290733456612, "entropy": 0.6504553481936455, "sharp": 0.11639169231057167} +{"time": "07:50:24", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3391.5927584658843, "loss": 0.6773724304512143, "entropy": 0.650679636746645, "sharp": 0.10639680176973343} +{"time": "07:51:21", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3393.1396837657157, "loss": 0.6611005216836929, "entropy": 0.6498031541705132, "sharp": 0.1116913054138422} +{"time": "07:52:18", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3396.25642456225, "loss": 0.6616475135087967, "entropy": 0.6504613123834133, "sharp": 0.11558049824088812} +{"time": "07:52:38", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.59246826171875, "noul_T": 0.720686105276827, "noul_auc": 0.7238021299218579, "noul_acc": 0.702, "noul_f1": 0.7682737169517885, "noul_nll": 0.5974437039494515, "noul_ece": 0.06587326323986054, "noul_brier": 0.20453869883011064, "noul_sharp": 0.145368292927742, "entropy": 0.6298882961273193} +{"time": "07:53:21", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9812951, "tok_per_s": 3324.467535176867, "rows_per_s": 5.70579453625253, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5789893269538879, "noul_T": 0.7367525983104597, "choice_ce": 0.6876552104949951, "choice_T": 1.0616362628517306, "score_ce": 1.0693273544311523, "score_T": 0.9881188547581379, "noul_auc": 0.7117098666666666, "noul_acc": 0.683, "noul_f1": 0.7643122676579925, "noul_nll": 0.5978014750629663, "noul_ece": 0.05898461356759074, "noul_brier": 0.2051665544090499, "noul_sharp": 0.1491762399673462, "entropy": 0.6274650692939758}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-M10-n16k-syn/DONE b/e28/S1-E28-M10-n16k-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..54833998880c486423c5963b6f03ad75b0982e9f --- /dev/null +++ b/e28/S1-E28-M10-n16k-syn/DONE @@ -0,0 +1 @@ +2026-10-20 10:17:10 \ No newline at end of file diff --git a/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..b45b01212d949ab3e4a392137119ddde38dffebc --- /dev/null +++ b/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "o_proj", + "v_proj", + "gate_proj", + "k_proj", + "up_proj", + "q_proj", + "down_proj", + "in_proj_qkv", + "out_proj", + "in_proj_z" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..90cb12bc138cc400819558dd4c8ecb9c45cb47b5 --- /dev/null +++ b/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:62e2f8d3a88d4ada04bb61d6ce161f835fc666951b8e301fb83af8b3fbbcaae4 +size 160486456 diff --git a/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d1220ef6c727472f5677764239334b3284ea4f2c --- /dev/null +++ b/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:384df5f614ed7f208e04003a7fb49666e84e8df0bcc2e71664ad5edd102c9e12 +size 176641832 diff --git a/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..c78ba7818b72db1e42cf460db1d3df3a4b214590 --- /dev/null +++ b/e28/S1-E28-M10-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.00029816626920364797, + -0.000298420520266518, + -0.0026337294839322567, + -0.001801400212571025, + -0.0006103252526372671, + 0.0018737963400781155, + 0.0013038155157119036, + -0.0022551037836819887, + -0.0005300171906128526, + -0.0005990835488773882, + -0.0003126576775684953, + 1.1003779945895076e-05, + -0.0008605930488556623, + -0.0036741262301802635, + -0.0045182351022958755, + -0.005161380860954523, + -0.005987720564007759, + -0.004427229519933462, + -0.004949240479618311, + -0.005030582193285227, + -0.003874005051329732, + -0.0034306200686842203, + -0.005743915680795908, + -0.0035071612801402807 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-M10-n16k-syn/ckpt-16000/calibration.json b/e28/S1-E28-M10-n16k-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..d85927ac0e7d43c08a4c975008a9f4872cd15ab1 --- /dev/null +++ b/e28/S1-E28-M10-n16k-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.7499735662738414, + "choice": 1.0068474373846998, + "score": 1.0045422922172058 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.7499735662738414, + "ce_before": 0.5744435787200928, + "ce_after": 0.568638026714325 + }, + "choice": { + "n": 60, + "T": 1.0068474373846998, + "ce_before": 0.6712058782577515, + "ce_after": 0.6711961627006531 + }, + "score": { + "n": 53, + "T": 1.0045422922172058, + "ce_before": 1.0601253509521484, + "ce_after": 1.0601203441619873 + } + }, + "source": "models/e28/S1-E28-M10-n16k-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-M10-n16k-syn/ckpt-16000/head.safetensors b/e28/S1-E28-M10-n16k-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..bac91fe49b42671917caa1f0cffae8830e017f17 --- /dev/null +++ b/e28/S1-E28-M10-n16k-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e29aded39a2b763e8dc8679f75766595dbbf240572ee567e465c97bd8f710646 +size 393464 diff --git a/e28/S1-E28-M10-n16k-syn/ckpt-16000/judge_config.json b/e28/S1-E28-M10-n16k-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..a5c4eca7ec1e703b4f56008201f4492facb15bc8 --- /dev/null +++ b/e28/S1-E28-M10-n16k-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-M10-n16k-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-M10-n16k", + "data_manifest": { + "arm": "E28-M10-n16k", + "name": null, + "train_sha256": "2ba17130e5e695ff4ca2f3aa2d668757227a4897d009f4aad7e92829fad2bcae", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9818658, + "tok_per_s": 3313.7081732868223, + "rows_per_s": 5.684022502726074, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5744435787200928, + "noul_T": 0.7499735662738414, + "choice_ce": 0.6712058782577515, + "choice_T": 1.0068474373846998, + "score_ce": 1.0601253509521484, + "score_T": 1.0045422922172058, + "noul_auc": 0.7213824, + "noul_acc": 0.687, + "noul_f1": 0.7703595011005135, + "noul_nll": 0.5918363256454467, + "noul_ece": 0.03898223993182183, + "noul_brier": 0.20276088802320869, + "noul_sharp": 0.15309229493141174, + "entropy": 0.6225576996803284 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-M10-n16k-syn/train_log.jsonl b/e28/S1-E28-M10-n16k-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..48f6c6806f69e7b6ab9b1aa4ce2590c127e13e93 --- /dev/null +++ b/e28/S1-E28-M10-n16k-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "09:27:46", "event": "start", "data": "data/e28/E28-M10-n16k", "data_manifest": {"arm": "E28-M10-n16k", "name": null, "train_sha256": "2ba17130e5e695ff4ca2f3aa2d668757227a4897d009f4aad7e92829fad2bcae", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.824934005737305} +{"time": "09:28:44", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3395.9434576669328, "loss": 0.6100487485527992, "entropy": 0.6578633710741997, "sharp": 0.11198114976286888} +{"time": "09:29:40", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3432.6592415893606, "loss": 0.7045050822198391, "entropy": 0.6172741055488586, "sharp": 0.1573221506550908} +{"time": "09:30:37", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3447.2518638112842, "loss": 0.6938936579972506, "entropy": 0.6440628916025162, "sharp": 0.14496233314275742} +{"time": "09:31:35", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3446.0068596841547, "loss": 0.6557534337043762, "entropy": 0.6702275648713112, "sharp": 0.09659549035131931} +{"time": "09:32:31", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3453.5410739528884, "loss": 0.6820793747901917, "entropy": 0.650994323194027, "sharp": 0.12144865654408932} +{"time": "09:33:25", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3450.8000793047872, "loss": 0.640854187309742, "entropy": 0.6424573957920074, "sharp": 0.11858639493584633} +{"time": "09:34:20", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3452.975767484592, "loss": 0.6121432557702065, "entropy": 0.6373509615659714, "sharp": 0.1404541414231062} +{"time": "09:35:16", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3455.8037154057174, "loss": 0.7084693126380444, "entropy": 0.6160237342119217, "sharp": 0.16685935482382774} +{"time": "09:36:12", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3459.5350085006057, "loss": 0.6386425942182541, "entropy": 0.6494212076067924, "sharp": 0.1202470138669014} +{"time": "09:37:09", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3464.680887446411, "loss": 0.6272236201912165, "entropy": 0.6410036459565163, "sharp": 0.12645962461829185} +{"time": "09:37:29", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.5956314206123352, "noul_T": 0.6984797400533347, "noul_auc": 0.7368820779396128, "noul_acc": 0.7, "noul_f1": 0.7641509433962265, "noul_nll": 0.5922642628848552, "noul_ece": 0.07741201174259185, "noul_brier": 0.2022020583721116, "noul_sharp": 0.1394229382276535, "entropy": 0.6347844004631042} +{"time": "09:38:26", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3356.734056643183, "loss": 0.6159855788573623, "entropy": 0.6409963071346283, "sharp": 0.13405491970479488} +{"time": "09:39:22", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3365.14399310511, "loss": 0.595206867903471, "entropy": 0.6000484153628349, "sharp": 0.1739121014252305} +{"time": "09:40:18", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3373.062049378464, "loss": 0.6032184660434723, "entropy": 0.6246584057807922, "sharp": 0.15253796614706516} +{"time": "09:41:14", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3380.954240865756, "loss": 0.6488698199391365, "entropy": 0.6392007246613503, "sharp": 0.13505388144403696} +{"time": "09:42:10", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3388.9751216766253, "loss": 0.6565089151263237, "entropy": 0.6586207672953606, "sharp": 0.11391774471849203} +{"time": "09:43:08", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3398.329451423392, "loss": 0.6917039826512337, "entropy": 0.6290627829730511, "sharp": 0.1504563670605421} +{"time": "09:44:05", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3403.763036240112, "loss": 0.7000130973756313, "entropy": 0.6526145339012146, "sharp": 0.1173095852136612} +{"time": "09:45:01", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3407.491266822165, "loss": 0.6259452328085899, "entropy": 0.6354352906346321, "sharp": 0.13545551989227533} +{"time": "09:45:56", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3411.7401979531273, "loss": 0.5817243345081806, "entropy": 0.5971872098743916, "sharp": 0.17717658169567585} +{"time": "09:46:53", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3416.2406227488373, "loss": 0.5766301527619362, "entropy": 0.633181132376194, "sharp": 0.14770043268799782} +{"time": "09:47:13", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5931195616722107, "noul_T": 0.6879390145389745, "noul_auc": 0.7325108436988405, "noul_acc": 0.71, "noul_f1": 0.779969650986343, "noul_nll": 0.5952536079287529, "noul_ece": 0.07498202133178711, "noul_brier": 0.20357922859091201, "noul_sharp": 0.13818307220935822, "entropy": 0.6340143084526062} +{"time": "09:48:11", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3363.1389383778533, "loss": 0.7324830964207649, "entropy": 0.6603128612041473, "sharp": 0.10196563135832548} +{"time": "09:49:09", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3366.4892135541418, "loss": 0.6070684026926756, "entropy": 0.6034758370369673, "sharp": 0.1789183709770441} +{"time": "09:50:05", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3370.0659984876256, "loss": 0.7011978551745415, "entropy": 0.6730314046144485, "sharp": 0.0834980858489871} +{"time": "09:51:00", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3373.94836369088, "loss": 0.6176299192011356, "entropy": 0.6388655975461006, "sharp": 0.1428214479237795} +{"time": "09:51:56", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3378.4957461696604, "loss": 0.6335353888571262, "entropy": 0.6278177425265312, "sharp": 0.14474612288177013} +{"time": "09:52:54", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3383.4265636149694, "loss": 0.6969103291630745, "entropy": 0.6501519680023193, "sharp": 0.1167315598577261} +{"time": "09:53:49", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3385.645233981425, "loss": 0.67652378231287, "entropy": 0.6603607088327408, "sharp": 0.10294342786073685} +{"time": "09:54:45", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3389.3031469716743, "loss": 0.710542457178235, "entropy": 0.6542830914258957, "sharp": 0.1206458043307066} +{"time": "09:55:41", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3392.3655148108655, "loss": 0.6637508608400822, "entropy": 0.6322581246495247, "sharp": 0.1387220099568367} +{"time": "09:56:37", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3395.294901227237, "loss": 0.6577721238136292, "entropy": 0.6471491977572441, "sharp": 0.12546446826308966} +{"time": "09:56:57", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5914047360420227, "noul_T": 0.7142428436228438, "noul_auc": 0.7337935224721945, "noul_acc": 0.69, "noul_f1": 0.7669172932330827, "noul_nll": 0.5934607314169407, "noul_ece": 0.07425392210483552, "noul_brier": 0.20282400597703348, "noul_sharp": 0.14581991732120514, "entropy": 0.6305782794952393} +{"time": "09:57:52", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3359.6732192563113, "loss": 0.6869237497448921, "entropy": 0.6447880156338215, "sharp": 0.13096304144710302} +{"time": "09:58:48", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3363.9772558646573, "loss": 0.6283286064863205, "entropy": 0.6557184234261513, "sharp": 0.10690976865589619} +{"time": "09:59:45", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3367.4101434942577, "loss": 0.6485901847481728, "entropy": 0.6473202407360077, "sharp": 0.1279229586943984} +{"time": "10:00:42", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3370.1590122746884, "loss": 0.7023767530918121, "entropy": 0.6597297117114067, "sharp": 0.10726573690772057} +{"time": "10:01:38", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3373.2301600732167, "loss": 0.6224467307329178, "entropy": 0.6221803165972233, "sharp": 0.1684778481721878} +{"time": "10:02:36", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3377.0119849960847, "loss": 0.6888734996318817, "entropy": 0.6651312932372093, "sharp": 0.09131032694131136} +{"time": "10:03:33", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3380.4862637059177, "loss": 0.6139164185151458, "entropy": 0.6279022824019194, "sharp": 0.1495692264288664} +{"time": "10:04:28", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3383.420155604395, "loss": 0.6542872041463852, "entropy": 0.6284296326339245, "sharp": 0.14589722454547882} +{"time": "10:05:25", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3387.4568459120105, "loss": 0.657096192240715, "entropy": 0.6556175872683525, "sharp": 0.11318864673376083} +{"time": "10:06:22", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3390.4411055722567, "loss": 0.6415223851799965, "entropy": 0.6533982381224632, "sharp": 0.11523059383034706} +{"time": "10:06:42", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5922847986221313, "noul_T": 0.6536423237818683, "noul_auc": 0.7361901064960928, "noul_acc": 0.7, "noul_f1": 0.7685185185185185, "noul_nll": 0.5941248633563518, "noul_ece": 0.06509018623828888, "noul_brier": 0.2031201667552924, "noul_sharp": 0.13586045801639557, "entropy": 0.6372879147529602} +{"time": "10:07:39", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3365.2004057236795, "loss": 0.6914893239736557, "entropy": 0.6546286381781101, "sharp": 0.11322860326617956} +{"time": "10:08:35", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3367.630356970402, "loss": 0.658904917538166, "entropy": 0.6574456989765167, "sharp": 0.11088940035551786} +{"time": "10:09:31", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3371.015554562461, "loss": 0.6366770900785923, "entropy": 0.6377274747937918, "sharp": 0.13475802168250084} +{"time": "10:10:28", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3373.0827605031113, "loss": 0.6748262271285057, "entropy": 0.6379967927932739, "sharp": 0.1309813503175974} +{"time": "10:11:25", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3375.331851017923, "loss": 0.611245671287179, "entropy": 0.6408358588814735, "sharp": 0.12345647253096104} +{"time": "10:12:21", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3377.0868717963335, "loss": 0.6320247650146484, "entropy": 0.6258900947868824, "sharp": 0.14644083753228188} +{"time": "10:13:18", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3379.2732971498517, "loss": 0.7263738997280598, "entropy": 0.6348090972751379, "sharp": 0.1452503204345703} +{"time": "10:14:13", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3381.501042963991, "loss": 0.6784677281975746, "entropy": 0.6459190994501114, "sharp": 0.12667710054665804} +{"time": "10:15:11", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3382.743220704638, "loss": 0.6158650293946266, "entropy": 0.6625316292047501, "sharp": 0.10314913466572762} +{"time": "10:16:07", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3385.006155101986, "loss": 0.6515998430550098, "entropy": 0.6372139006853104, "sharp": 0.13457430619746447} +{"time": "10:16:27", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5881441235542297, "noul_T": 0.732185747603024, "noul_auc": 0.7347892862567721, "noul_acc": 0.7, "noul_f1": 0.7713414634146342, "noul_nll": 0.5904512218534946, "noul_ece": 0.06275870645046236, "noul_brier": 0.20170839538402627, "noul_sharp": 0.1487353891134262, "entropy": 0.6255123019218445} +{"time": "10:17:10", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9818658, "tok_per_s": 3313.7081732868223, "rows_per_s": 5.684022502726074, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5744435787200928, "noul_T": 0.7499735662738414, "choice_ce": 0.6712058782577515, "choice_T": 1.0068474373846998, "score_ce": 1.0601253509521484, "score_T": 1.0045422922172058, "noul_auc": 0.7213824, "noul_acc": 0.687, "noul_f1": 0.7703595011005135, "noul_nll": 0.5918363256454467, "noul_ece": 0.03898223993182183, "noul_brier": 0.20276088802320869, "noul_sharp": 0.15309229493141174, "entropy": 0.6225576996803284}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-M10P1-n16k-syn/DONE b/e28/S1-E28-M10P1-n16k-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..4938e5340e9c71e43f900b7f93c3970660b6bce1 --- /dev/null +++ b/e28/S1-E28-M10P1-n16k-syn/DONE @@ -0,0 +1 @@ +2026-10-20 16:21:13 \ No newline at end of file diff --git a/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2967ae8904b748b320478ae35e07ad6d0d257cde --- /dev/null +++ b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "v_proj", + "k_proj", + "out_proj", + "gate_proj", + "in_proj_qkv", + "up_proj", + "q_proj", + "in_proj_z", + "down_proj", + "o_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5400e7b8703ea348650433ee433bd1bef5f21736 --- /dev/null +++ b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b9ff0d28c5fa4bb9572e125ffdd798a06f950097a419815259c252f1fae9b6a3 +size 160486456 diff --git a/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..eec6818ec6e253a82132668a657d2ef7fa867558 --- /dev/null +++ b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77e8ad6b9872e184d3219323d8812855d09fc73c918ed2f7a963fb3e0405637f +size 176641832 diff --git a/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..475cdcb0c7deec2d37a8b473770ee0516dfb4b11 --- /dev/null +++ b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.000409875123295933, + -0.0004101314989384264, + -0.0021534895058721304, + -0.001860003569163382, + -0.0008476248476654291, + 0.0019733395893126726, + 0.0016671096673235297, + -0.0025478086899966, + -0.00028338219271972775, + -0.00028915764414705336, + -0.0001285076723434031, + 0.0008073451463133097, + -0.0005743084475398064, + -0.0036162924952805042, + -0.004263429902493954, + -0.005093412473797798, + -0.005931205581873655, + -0.004397733602672815, + -0.004943625535815954, + -0.004997676704078913, + -0.0038396355230361223, + -0.0034048345405608416, + -0.005696979351341724, + -0.0034947521053254604 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/calibration.json b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..111d6cff306c93a594175957d02ea2308f52a5b7 --- /dev/null +++ b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.7769211806295386, + "choice": 1.0554360712007262, + "score": 0.9861511896147653 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.7769211806295386, + "ce_before": 0.5735545754432678, + "ce_after": 0.5689408183097839 + }, + "choice": { + "n": 60, + "T": 1.0554360712007262, + "ce_before": 0.6860017776489258, + "ce_after": 0.6853616237640381 + }, + "score": { + "n": 53, + "T": 0.9861511896147653, + "ce_before": 1.0551782846450806, + "ce_after": 1.0551178455352783 + } + }, + "source": "models/e28/S1-E28-M10P1-n16k-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/head.safetensors b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..509f85f8997dcf7ef619fcc00d19bfb06304817f --- /dev/null +++ b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:faaf3dd8fccd256437b7b10cefb822f5f9f1222f1cce1882c9a85322c840f9fb +size 393464 diff --git a/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/judge_config.json b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..950981fd17283fc88550be9aab6af39964c7ca92 --- /dev/null +++ b/e28/S1-E28-M10P1-n16k-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-M10P1-n16k-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-M10P1-n16k", + "data_manifest": { + "arm": "E28-M10P1-n16k", + "name": null, + "train_sha256": "7be246e93c2588f776a5f89c002d902a516bab65bc904fdc3db7ff512bc147ce", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9865291, + "tok_per_s": 3330.1849964151397, + "rows_per_s": 5.685283451749966, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5735545754432678, + "noul_T": 0.7769211806295386, + "choice_ce": 0.6860017776489258, + "choice_T": 1.0554360712007262, + "score_ce": 1.0551782846450806, + "score_T": 0.9861511896147653, + "noul_auc": 0.7191552, + "noul_acc": 0.698, + "noul_f1": 0.7762962962962963, + "noul_nll": 0.5914556209370494, + "noul_ece": 0.048010656148195274, + "noul_brier": 0.2024473024458909, + "noul_sharp": 0.1621304303407669, + "entropy": 0.6183435916900635 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-M10P1-n16k-syn/train_log.jsonl b/e28/S1-E28-M10P1-n16k-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..60802156ac7076f7d1f488562afc4aa9dc161a34 --- /dev/null +++ b/e28/S1-E28-M10P1-n16k-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "15:31:51", "event": "start", "data": "data/e28/E28-M10P1-n16k", "data_manifest": {"arm": "E28-M10P1-n16k", "name": null, "train_sha256": "7be246e93c2588f776a5f89c002d902a516bab65bc904fdc3db7ff512bc147ce", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.800007343292236} +{"time": "15:32:48", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3413.18969493185, "loss": 0.5770136564970016, "entropy": 0.6013723723590374, "sharp": 0.19633070193231106} +{"time": "15:33:44", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3460.533379272013, "loss": 0.6829772666096687, "entropy": 0.6553364358842373, "sharp": 0.11164396163076162} +{"time": "15:34:42", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3475.577873266482, "loss": 0.6583068333566189, "entropy": 0.6647536754608154, "sharp": 0.09844676498323679} +{"time": "15:35:38", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3478.765433973951, "loss": 0.6919600255787373, "entropy": 0.6636394299566746, "sharp": 0.09736234787851572} +{"time": "15:36:33", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3478.407651338676, "loss": 0.670541949570179, "entropy": 0.6513997092843056, "sharp": 0.12489207740873098} +{"time": "15:37:29", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3482.815442477279, "loss": 0.6420967876911163, "entropy": 0.6711881682276726, "sharp": 0.09153465367853642} +{"time": "15:38:26", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3485.1908018742774, "loss": 0.5863455515354872, "entropy": 0.622340053319931, "sharp": 0.16095477156341076} +{"time": "15:39:24", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3486.5524902422585, "loss": 0.6621488481760025, "entropy": 0.6520817875862122, "sharp": 0.1218183096498251} +{"time": "15:40:19", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3488.331782940861, "loss": 0.6836168244481087, "entropy": 0.664107371121645, "sharp": 0.10315894614905119} +{"time": "15:41:15", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3485.1816031550416, "loss": 0.649325555190444, "entropy": 0.6482464969158173, "sharp": 0.13002939522266388} +{"time": "15:41:35", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.5960443615913391, "noul_T": 0.7224970159103625, "noul_auc": 0.7252198275134598, "noul_acc": 0.69, "noul_f1": 0.7723935389133627, "noul_nll": 0.5982213671803475, "noul_ece": 0.07068759310245513, "noul_brier": 0.20477424771314173, "noul_sharp": 0.1453370898962021, "entropy": 0.6330079436302185} +{"time": "15:42:30", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3380.0825239788683, "loss": 0.6682743430137634, "entropy": 0.6387989073991776, "sharp": 0.13733062148094177} +{"time": "15:43:26", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3385.283057820895, "loss": 0.6369949597865343, "entropy": 0.664315789937973, "sharp": 0.09187121409922838} +{"time": "15:44:22", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3392.1199144335187, "loss": 0.6194523721933365, "entropy": 0.619594968855381, "sharp": 0.1682006847113371} +{"time": "15:45:20", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3399.3257137941664, "loss": 0.6384348049759865, "entropy": 0.6611604690551758, "sharp": 0.1065277848392725} +{"time": "15:46:15", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3404.538846911732, "loss": 0.6592436209321022, "entropy": 0.6531195566058159, "sharp": 0.11518852971494198} +{"time": "15:47:12", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3410.904948655331, "loss": 0.7061475366353989, "entropy": 0.658478707075119, "sharp": 0.09994554053992033} +{"time": "15:48:07", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3415.318659250689, "loss": 0.6632629185914993, "entropy": 0.6729573793709278, "sharp": 0.08020197879523039} +{"time": "15:49:03", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3421.508288918258, "loss": 0.6117471121251583, "entropy": 0.634588036686182, "sharp": 0.14776423294097185} +{"time": "15:49:58", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3425.3660056455515, "loss": 0.634715598076582, "entropy": 0.6508324667811394, "sharp": 0.13239708822220564} +{"time": "15:50:54", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3429.084936444736, "loss": 0.6466965302824974, "entropy": 0.6282579228281975, "sharp": 0.15463204495608807} +{"time": "15:51:14", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5868313312530518, "noul_T": 0.7957678124629807, "noul_auc": 0.7371521155761085, "noul_acc": 0.706, "noul_f1": 0.7776096822995462, "noul_nll": 0.5903388968408108, "noul_ece": 0.0694612179994583, "noul_brier": 0.20156302005543797, "noul_sharp": 0.1536959409713745, "entropy": 0.6159988641738892} +{"time": "15:52:10", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3377.5846126686242, "loss": 0.6573365703225136, "entropy": 0.6669176891446114, "sharp": 0.09173294436186552} +{"time": "15:53:07", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3382.8484525248828, "loss": 0.625731460750103, "entropy": 0.6380208991467953, "sharp": 0.13441915903240442} +{"time": "15:54:05", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3387.158523157695, "loss": 0.6537837907671928, "entropy": 0.6311176419258118, "sharp": 0.15432934742420912} +{"time": "15:55:03", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3391.7269509968805, "loss": 0.6569702737033367, "entropy": 0.6546890698373318, "sharp": 0.11343663465231657} +{"time": "15:56:00", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3396.566872418787, "loss": 0.5975664034485817, "entropy": 0.6460825502872467, "sharp": 0.12390232738107443} +{"time": "15:56:55", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3400.014158663088, "loss": 0.6100738421082497, "entropy": 0.6430663093924522, "sharp": 0.1326448107138276} +{"time": "15:57:51", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3403.2008390874776, "loss": 0.7068752944469452, "entropy": 0.6296126171946526, "sharp": 0.14069452974945307} +{"time": "15:58:46", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3408.3930601238008, "loss": 0.6571051925420761, "entropy": 0.6546614319086075, "sharp": 0.11463713645935059} +{"time": "15:59:41", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3410.961779105179, "loss": 0.7003757990896702, "entropy": 0.632878340780735, "sharp": 0.15215817652642727} +{"time": "16:00:39", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3413.6209321808897, "loss": 0.668920248746872, "entropy": 0.6700474172830582, "sharp": 0.08504971023648977} +{"time": "16:00:59", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.6158488988876343, "noul_T": 0.8152103731798707, "noul_auc": 0.726249346002599, "noul_acc": 0.682, "noul_f1": 0.7195767195767195, "noul_nll": 0.6157990889400243, "noul_ece": 0.10593590170145034, "noul_brier": 0.21312713047170426, "noul_sharp": 0.14067211747169495, "entropy": 0.6369451880455017} +{"time": "16:01:57", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3379.110245365357, "loss": 0.6192936860024929, "entropy": 0.639333214610815, "sharp": 0.1326094539836049} +{"time": "16:02:52", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3382.2007088970377, "loss": 0.6304379552602768, "entropy": 0.6492158696055412, "sharp": 0.11933540739119053} +{"time": "16:03:49", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3386.4656213353555, "loss": 0.6708593815565109, "entropy": 0.6691597551107407, "sharp": 0.08842133451253176} +{"time": "16:04:47", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3390.4654731289775, "loss": 0.6764287576079369, "entropy": 0.6570648178458214, "sharp": 0.11048906203359365} +{"time": "16:05:42", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3393.0236436894966, "loss": 0.6111768409609795, "entropy": 0.641745038330555, "sharp": 0.13162633031606674} +{"time": "16:06:39", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3395.3312570413436, "loss": 0.7014862596988678, "entropy": 0.6586485765874386, "sharp": 0.10697954054921865} +{"time": "16:07:37", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3398.92927322851, "loss": 0.6040123999118805, "entropy": 0.6082264520227909, "sharp": 0.18517939932644367} +{"time": "16:08:36", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3400.5972864662867, "loss": 0.6495504342019558, "entropy": 0.65518444404006, "sharp": 0.1133860033005476} +{"time": "16:09:33", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3403.8189753848665, "loss": 0.7093628719449043, "entropy": 0.6537637636065483, "sharp": 0.11518269870430231} +{"time": "16:10:29", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3407.345581084016, "loss": 0.6282716318964958, "entropy": 0.6509821563959122, "sharp": 0.12243939097970724} +{"time": "16:10:49", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5912967324256897, "noul_T": 0.7576067121930582, "noul_auc": 0.7270425815598048, "noul_acc": 0.712, "noul_f1": 0.7757009345794392, "noul_nll": 0.5935759347528219, "noul_ece": 0.06600151920318603, "noul_brier": 0.20261482688820806, "noul_sharp": 0.1535715013742447, "entropy": 0.624973714351654} +{"time": "16:11:45", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3381.898593516214, "loss": 0.6442261002957821, "entropy": 0.6235337536782026, "sharp": 0.1496837278828025} +{"time": "16:12:42", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3384.632062135497, "loss": 0.6366482563316822, "entropy": 0.6566184684634209, "sharp": 0.10810046456754208} +{"time": "16:13:38", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3386.9133573517493, "loss": 0.6827273853123188, "entropy": 0.6373638529330492, "sharp": 0.1423604991286993} +{"time": "16:14:34", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3388.7804047414816, "loss": 0.6264001727104187, "entropy": 0.6314874291419983, "sharp": 0.14877322129905224} +{"time": "16:15:31", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3391.552102510729, "loss": 0.6527801752090454, "entropy": 0.6405499503016472, "sharp": 0.1344473399221897} +{"time": "16:16:26", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3394.0545226734052, "loss": 0.6626713071018457, "entropy": 0.6402172185480595, "sharp": 0.13957634381949902} +{"time": "16:17:23", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3396.381167725961, "loss": 0.7259967923164368, "entropy": 0.6385600827634335, "sharp": 0.1356890108436346} +{"time": "16:18:20", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3398.043512050349, "loss": 0.6306109204888344, "entropy": 0.6466524004936218, "sharp": 0.128496746532619} +{"time": "16:19:16", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3400.225138395327, "loss": 0.6264664381742477, "entropy": 0.6411445997655392, "sharp": 0.13463375717401505} +{"time": "16:20:11", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3401.724897800272, "loss": 0.6832641996443272, "entropy": 0.6528444737195969, "sharp": 0.1168731776997447} +{"time": "16:20:30", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5877655148506165, "noul_T": 0.7549100218006332, "noul_auc": 0.7320214004826922, "noul_acc": 0.716, "noul_f1": 0.7795031055900621, "noul_nll": 0.589799154639244, "noul_ece": 0.06508569729328156, "noul_brier": 0.20100635894807675, "noul_sharp": 0.15704989433288574, "entropy": 0.6230216026306152} +{"time": "16:21:13", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9865291, "tok_per_s": 3330.1849964151397, "rows_per_s": 5.685283451749966, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5735545754432678, "noul_T": 0.7769211806295386, "choice_ce": 0.6860017776489258, "choice_T": 1.0554360712007262, "score_ce": 1.0551782846450806, "score_T": 0.9861511896147653, "noul_auc": 0.7191552, "noul_acc": 0.698, "noul_f1": 0.7762962962962963, "noul_nll": 0.5914556209370494, "noul_ece": 0.048010656148195274, "noul_brier": 0.2024473024458909, "noul_sharp": 0.1621304303407669, "entropy": 0.6183435916900635}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-M10P40-n16k-syn/DONE b/e28/S1-E28-M10P40-n16k-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..6911a2fb054a39c67e3571bd0cee1c4b63d33877 --- /dev/null +++ b/e28/S1-E28-M10P40-n16k-syn/DONE @@ -0,0 +1 @@ +2026-10-20 17:34:24 \ No newline at end of file diff --git a/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..11b92dd82f32de67d328aaba5ab4d7d70ad9d4e2 --- /dev/null +++ b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "v_proj", + "gate_proj", + "o_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "q_proj", + "up_proj", + "down_proj", + "out_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dee1be33a0ce40a2f7a44eae7b2ea28b74a2759b --- /dev/null +++ b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c0799b009ce1825f5692e6537a396d8631c1e5ea3bf149ed29e80724fc698dc8 +size 160486456 diff --git a/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..07c2089c59a54da04963ea5150c059b5c65d9ff6 --- /dev/null +++ b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45b831d33cf26b9fa544c0ac263fdde556994e6c39108ea528b95b4e51ca40f0 +size 176641832 diff --git a/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..fe0b7bf78f1e16e0ae973f493b6f4d077cf6e752 --- /dev/null +++ b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.00029545993311330676, + -0.0002957115648314357, + -0.002514040097594261, + -0.0016292216023430228, + -0.000663517217617482, + 0.0021830580662935972, + 0.0018118764273822308, + -0.001836547744460404, + -0.0005211382522247732, + -0.00018114523845724761, + -3.332009873702191e-05, + 0.00030125127523206174, + -0.00034467881778255105, + -0.0034366166219115257, + -0.004174028057605028, + -0.005128026939928532, + -0.005891313776373863, + -0.0043664840050041676, + -0.004935290198773146, + -0.004985618870705366, + -0.003741042921319604, + -0.003376793349161744, + -0.005712005775421858, + -0.00347433565184474 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/calibration.json b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..4978080ecc544fbbe4403d305e0204a5e89b426f --- /dev/null +++ b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.7936163840258926, + "choice": 1.0373980355006043, + "score": 0.9535006548394708 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.7936163840258926, + "ce_before": 0.5721228122711182, + "ce_after": 0.5681952238082886 + }, + "choice": { + "n": 60, + "T": 1.0373980355006043, + "ce_before": 0.67246413230896, + "ce_after": 0.6721594333648682 + }, + "score": { + "n": 53, + "T": 0.9535006548394708, + "ce_before": 1.06044602394104, + "ce_after": 1.0597749948501587 + } + }, + "source": "models/e28/S1-E28-M10P40-n16k-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/head.safetensors b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c62047c15add18cbb9957d28540b89d84e0967eb --- /dev/null +++ b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:67fe79afda64de0cb6bdca7b9c1c8d230f172fa9f9d38284d8ca64d6f27e826b +size 393464 diff --git a/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/judge_config.json b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..aba8d51e35d798778d743cd9debb1a5799d0d292 --- /dev/null +++ b/e28/S1-E28-M10P40-n16k-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-M10P40-n16k-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-M10P40-n16k", + "data_manifest": { + "arm": "E28-M10P40-n16k", + "name": null, + "train_sha256": "39574ddaa158a99572227f29fcc2be761352b45f2f70805ff1c7883379ef2228", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9820167, + "tok_per_s": 3325.19537240611, + "rows_per_s": 5.702850107377566, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5721228122711182, + "noul_T": 0.7936163840258926, + "choice_ce": 0.67246413230896, + "choice_T": 1.0373980355006043, + "score_ce": 1.06044602394104, + "score_T": 0.9535006548394708, + "noul_auc": 0.7226837333333334, + "noul_acc": 0.699, + "noul_f1": 0.773172569706104, + "noul_nll": 0.5896362618133426, + "noul_ece": 0.051595462948083884, + "noul_brier": 0.20191176004985892, + "noul_sharp": 0.1634892374277115, + "entropy": 0.6141766905784607 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-M10P40-n16k-syn/train_log.jsonl b/e28/S1-E28-M10P40-n16k-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..7d3f02a156ff90cab2ea22f36276cda5073b5644 --- /dev/null +++ b/e28/S1-E28-M10P40-n16k-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "16:45:10", "event": "start", "data": "data/e28/E28-M10P40-n16k", "data_manifest": {"arm": "E28-M10P40-n16k", "name": null, "train_sha256": "39574ddaa158a99572227f29fcc2be761352b45f2f70805ff1c7883379ef2228", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.638131380081177} +{"time": "16:46:07", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3455.701937308954, "loss": 0.6403607912361622, "entropy": 0.6541971042752266, "sharp": 0.11336817312985659} +{"time": "16:47:04", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3483.657241530826, "loss": 0.7572037354111671, "entropy": 0.5939787998795509, "sharp": 0.18362688180059195} +{"time": "16:48:01", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3483.3313898711895, "loss": 0.6978725977241993, "entropy": 0.5995655320584774, "sharp": 0.1900278590619564} +{"time": "16:48:56", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3482.831806277697, "loss": 0.7036403901875019, "entropy": 0.6572661474347115, "sharp": 0.1069600684568286} +{"time": "16:49:51", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3479.9237853729437, "loss": 0.6372505947947502, "entropy": 0.6219921708106995, "sharp": 0.16206148825585842} +{"time": "16:50:47", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3493.1712744304577, "loss": 0.5845348797738552, "entropy": 0.6367757730185986, "sharp": 0.133145185187459} +{"time": "16:51:42", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3494.8627701187233, "loss": 0.6319153904914856, "entropy": 0.6818341389298439, "sharp": 0.05842152051627636} +{"time": "16:52:39", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3490.3486605945263, "loss": 0.6729952953755856, "entropy": 0.6243966817855835, "sharp": 0.15194333717226982} +{"time": "16:53:34", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3489.656175346252, "loss": 0.622497908771038, "entropy": 0.6426990777254105, "sharp": 0.1182904476299882} +{"time": "16:54:31", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3493.546723566235, "loss": 0.6625543385744095, "entropy": 0.6255499068647623, "sharp": 0.15097005665302277} +{"time": "16:54:51", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.6056156158447266, "noul_T": 0.683613507874773, "noul_auc": 0.7312619196300484, "noul_acc": 0.708, "noul_f1": 0.7827380952380952, "noul_nll": 0.6039456658661365, "noul_ece": 0.08665356969833377, "noul_brier": 0.20694239045899823, "noul_sharp": 0.1299819052219391, "entropy": 0.6422101259231567} +{"time": "16:55:48", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3381.9086968620854, "loss": 0.6446554400026798, "entropy": 0.6498633176088333, "sharp": 0.12302601709961891} +{"time": "16:56:42", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3389.2243427696285, "loss": 0.6421851385384798, "entropy": 0.6602730490267277, "sharp": 0.09480368159711361} +{"time": "16:57:36", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3396.0455651412135, "loss": 0.6387088969349861, "entropy": 0.6130596175789833, "sharp": 0.16664594318717718} +{"time": "16:58:32", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3404.8357812346558, "loss": 0.6025540083646774, "entropy": 0.6425704471766949, "sharp": 0.12057922873646021} +{"time": "16:59:28", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3409.9071778421862, "loss": 0.6151242665946484, "entropy": 0.6350378151983023, "sharp": 0.15412558056414127} +{"time": "17:00:24", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3414.7486285234927, "loss": 0.6716683506965637, "entropy": 0.5761628840118647, "sharp": 0.20104592852294445} +{"time": "17:01:19", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3419.213771288617, "loss": 0.5964947044849396, "entropy": 0.6494682244956493, "sharp": 0.12307983729988337} +{"time": "17:02:15", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3423.396634290794, "loss": 0.6177096739411354, "entropy": 0.6711957454681396, "sharp": 0.08147689327597618} +{"time": "17:03:11", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3426.120546077965, "loss": 0.658025580458343, "entropy": 0.6541718803346157, "sharp": 0.11983995977789164} +{"time": "17:04:08", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3430.377301488183, "loss": 0.6141386777162552, "entropy": 0.6295526623725891, "sharp": 0.1472106035798788} +{"time": "17:04:28", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5898134708404541, "noul_T": 0.9617091660304182, "noul_auc": 0.7375402946785709, "noul_acc": 0.7, "noul_f1": 0.7819767441860465, "noul_nll": 0.5848946662843227, "noul_ece": 0.06847818994522094, "noul_brier": 0.19912294724357898, "noul_sharp": 0.18461810052394867, "entropy": 0.5960484147071838} +{"time": "17:05:25", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3377.948569676899, "loss": 0.6613380564376712, "entropy": 0.6486588902771473, "sharp": 0.12644754722714424} +{"time": "17:06:21", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3381.798994029936, "loss": 0.6384977549314499, "entropy": 0.6398581862449646, "sharp": 0.13425691053271294} +{"time": "17:07:18", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3384.4780165978223, "loss": 0.6378122828900814, "entropy": 0.6505446508526802, "sharp": 0.11962900776416063} +{"time": "17:08:15", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3388.6196762725854, "loss": 0.6598417572677135, "entropy": 0.6403933800756931, "sharp": 0.14041622076183558} +{"time": "17:09:12", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3393.9342869132893, "loss": 0.6357480324804783, "entropy": 0.6285230740904808, "sharp": 0.1576135102659464} +{"time": "17:10:08", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3397.976874309936, "loss": 0.62338537722826, "entropy": 0.652957808226347, "sharp": 0.1207507373765111} +{"time": "17:11:03", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3401.2059680194566, "loss": 0.6495342738926411, "entropy": 0.6656588315963745, "sharp": 0.09409788157790899} +{"time": "17:11:59", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3403.6523742591967, "loss": 0.6432218514382839, "entropy": 0.642509188503027, "sharp": 0.11506645940244198} +{"time": "17:12:54", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3406.011694826732, "loss": 0.6637626215815544, "entropy": 0.6559516452252865, "sharp": 0.11204951256513596} +{"time": "17:13:51", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3410.3710723526287, "loss": 0.6642355397343636, "entropy": 0.6417679227888584, "sharp": 0.13322205189615488} +{"time": "17:14:11", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5869750380516052, "noul_T": 0.8809182329941555, "noul_auc": 0.7371521155761085, "noul_acc": 0.682, "noul_f1": 0.7788595271210014, "noul_nll": 0.590536227196455, "noul_ece": 0.06798189330101014, "noul_brier": 0.20235864047072763, "noul_sharp": 0.16915255784988403, "entropy": 0.6051074266433716} +{"time": "17:15:07", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3374.017874294268, "loss": 0.6049460396170616, "entropy": 0.6192969605326653, "sharp": 0.15581472590565681} +{"time": "17:16:03", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3378.2337715273106, "loss": 0.6658960841596127, "entropy": 0.6582859307527542, "sharp": 0.10285252146422863} +{"time": "17:16:59", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3380.9834773665325, "loss": 0.6309775188565254, "entropy": 0.6360249146819115, "sharp": 0.13989773578941822} +{"time": "17:17:54", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3385.1892884465624, "loss": 0.6597106633707881, "entropy": 0.6175982104614377, "sharp": 0.1645418917760253} +{"time": "17:18:51", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3389.313879860058, "loss": 0.6557650789618492, "entropy": 0.6463417373597622, "sharp": 0.1198677085340023} +{"time": "17:19:46", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3392.399165226995, "loss": 0.6346933282911777, "entropy": 0.6376127973198891, "sharp": 0.12364564370363951} +{"time": "17:20:43", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3395.365824095542, "loss": 0.6899731382727623, "entropy": 0.6309513188898563, "sharp": 0.1391269899904728} +{"time": "17:21:37", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3397.469490282424, "loss": 0.6274153850972652, "entropy": 0.6405365914106369, "sharp": 0.12904574815183878} +{"time": "17:22:34", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3399.6297130688586, "loss": 0.6861204942688346, "entropy": 0.6503848508000374, "sharp": 0.10543647687882185} +{"time": "17:23:32", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3403.001781662534, "loss": 0.6548432037234306, "entropy": 0.6370681449770927, "sharp": 0.13254490867257118} +{"time": "17:23:52", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5863056778907776, "noul_T": 0.8149615780023207, "noul_auc": 0.7332028151423604, "noul_acc": 0.706, "noul_f1": 0.776255707762557, "noul_nll": 0.5886026494950056, "noul_ece": 0.06089252924919128, "noul_brier": 0.20094221131483445, "noul_sharp": 0.16097168624401093, "entropy": 0.6133525967597961} +{"time": "17:24:49", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3376.879852282775, "loss": 0.6281008310616016, "entropy": 0.6639142651110888, "sharp": 0.09336040262132883} +{"time": "17:25:47", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3379.9662636416797, "loss": 0.5983796566724777, "entropy": 0.6264648586511612, "sharp": 0.1501619853079319} +{"time": "17:26:45", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3382.1615866035786, "loss": 0.6393643319606781, "entropy": 0.6166650075465441, "sharp": 0.15496844984591007} +{"time": "17:27:41", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3385.4727540085023, "loss": 0.6451219841837883, "entropy": 0.6437481716275215, "sharp": 0.12111335340887308} +{"time": "17:28:39", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3387.816634581529, "loss": 0.6220843717455864, "entropy": 0.6673966348171234, "sharp": 0.09127752389758825} +{"time": "17:29:35", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3389.5503901312536, "loss": 0.6465622503310442, "entropy": 0.6569411773234606, "sharp": 0.11052277591079473} +{"time": "17:30:32", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3392.0329652411992, "loss": 0.6515473760664463, "entropy": 0.6541538089513779, "sharp": 0.11867711134254932} +{"time": "17:31:28", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3393.12863023791, "loss": 0.6516725607216358, "entropy": 0.6409944966435432, "sharp": 0.13743835873901844} +{"time": "17:32:25", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3395.572191861588, "loss": 0.6411801427602768, "entropy": 0.6517010554671288, "sharp": 0.11329328920692205} +{"time": "17:33:21", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3396.9271301715826, "loss": 0.6412891829386353, "entropy": 0.6495162472128868, "sharp": 0.1254839738830924} +{"time": "17:33:41", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.587288498878479, "noul_T": 0.7755967648258878, "noul_auc": 0.735582521813978, "noul_acc": 0.712, "noul_f1": 0.7714285714285715, "noul_nll": 0.5887580643296242, "noul_ece": 0.06437312817573548, "noul_brier": 0.20094398217515708, "noul_sharp": 0.15758994221687317, "entropy": 0.6195856928825378} +{"time": "17:34:24", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9820167, "tok_per_s": 3325.19537240611, "rows_per_s": 5.702850107377566, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5721228122711182, "noul_T": 0.7936163840258926, "choice_ce": 0.67246413230896, "choice_T": 1.0373980355006043, "score_ce": 1.06044602394104, "score_T": 0.9535006548394708, "noul_auc": 0.7226837333333334, "noul_acc": 0.699, "noul_f1": 0.773172569706104, "noul_nll": 0.5896362618133426, "noul_ece": 0.051595462948083884, "noul_brier": 0.20191176004985892, "noul_sharp": 0.1634892374277115, "entropy": 0.6141766905784607}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-M1P1-n16k-syn/DONE b/e28/S1-E28-M1P1-n16k-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..33b81b67d57d43a8db1723f83e379a420a3daf1e --- /dev/null +++ b/e28/S1-E28-M1P1-n16k-syn/DONE @@ -0,0 +1 @@ +2026-10-20 13:55:16 \ No newline at end of file diff --git a/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..61ab45412e3ea221633dc0b7b9bdb5db6307ac96 --- /dev/null +++ b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "v_proj", + "in_proj_qkv", + "q_proj", + "down_proj", + "in_proj_z", + "gate_proj", + "o_proj", + "out_proj", + "up_proj", + "k_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..beb737832de80968ccf83eff014ab172c9a28b5e --- /dev/null +++ b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3cfb9617fdfabf666758de657a30d3822bf6b2741e255eaf1819e3c711269ac6 +size 160486456 diff --git a/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..31a127825cbdf2cf821b5dec58ea08ccd8023de1 --- /dev/null +++ b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:310793221105809096145340fb8b7c68d592714e91fa7553a55dbec7d35525e8 +size 176641832 diff --git a/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..2db25f462c808a3cb0b7aeaa805dd9c4700ab6d6 --- /dev/null +++ b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.0002981233992613852, + -0.0002983761951327324, + -0.002645886270329356, + -0.0016880007460713387, + -0.0007843613275326788, + 0.0019988061394542456, + 0.001451568678021431, + -0.0025670710019767284, + -0.00039818824734538794, + -0.00041181326378136873, + -1.417321982444264e-05, + 0.00018017432012129575, + -0.0008135371608659625, + -0.0034676319919526577, + -0.004418123513460159, + -0.0051560248248279095, + -0.006017521023750305, + -0.004424345679581165, + -0.004941635765135288, + -0.004991940222680569, + -0.0037378983106464148, + -0.0033777246717363596, + -0.005773702636361122, + -0.003484088694676757 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/calibration.json b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..23d69ac1e8fc5b094e03ea52334809c33b4193b6 --- /dev/null +++ b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.8128870384220848, + "choice": 0.9823765467076169, + "score": 0.9929303222249871 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.8128870384220848, + "ce_before": 0.5800862908363342, + "ce_after": 0.5770651698112488 + }, + "choice": { + "n": 60, + "T": 0.9823765467076169, + "ce_before": 0.668471097946167, + "ce_after": 0.6684014201164246 + }, + "score": { + "n": 53, + "T": 0.9929303222249871, + "ce_before": 1.0641282796859741, + "ce_after": 1.0641119480133057 + } + }, + "source": "models/e28/S1-E28-M1P1-n16k-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/head.safetensors b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9e2da6afc0a6340f3afbbfb0d00703e612f652c4 --- /dev/null +++ b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4ad9eb99af5376e886ef355073355b27789de889e95f59fcfadd22add792e881 +size 393464 diff --git a/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/judge_config.json b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..6cad24b8ec02008385af5f240e61a909883d1483 --- /dev/null +++ b/e28/S1-E28-M1P1-n16k-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-M1P1-n16k-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-M1P1-n16k", + "data_manifest": { + "arm": "E28-M1P1-n16k", + "name": null, + "train_sha256": "01645a38ea82ed667cb692ff3dcaf6d2bdc5936e9be7e7ba5e3f23386bef3f07", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9861110, + "tok_per_s": 3316.668544463796, + "rows_per_s": 5.664608911729549, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5800862908363342, + "noul_T": 0.8128870384220848, + "choice_ce": 0.668471097946167, + "choice_T": 0.9823765467076169, + "score_ce": 1.0641282796859741, + "score_T": 0.9929303222249871, + "noul_auc": 0.7086250666666667, + "noul_acc": 0.689, + "noul_f1": 0.7663410969196093, + "noul_nll": 0.5972189504876733, + "noul_ece": 0.041735121816396714, + "noul_brier": 0.20542506687095102, + "noul_sharp": 0.15873706340789795, + "entropy": 0.6186158657073975 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-M1P1-n16k-syn/train_log.jsonl b/e28/S1-E28-M1P1-n16k-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..06154a781a68a39ae4e9e1875b5d680b47e62366 --- /dev/null +++ b/e28/S1-E28-M1P1-n16k-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "13:05:43", "event": "start", "data": "data/e28/E28-M1P1-n16k", "data_manifest": {"arm": "E28-M1P1-n16k", "name": null, "train_sha256": "01645a38ea82ed667cb692ff3dcaf6d2bdc5936e9be7e7ba5e3f23386bef3f07", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.77591347694397} +{"time": "13:06:41", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3428.6902609920603, "loss": 0.6499541215598583, "entropy": 0.6815319582819939, "sharp": 0.05769501905888319} +{"time": "13:07:37", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3448.1572533463627, "loss": 0.6885519549250603, "entropy": 0.6455572098493576, "sharp": 0.14128214307129383} +{"time": "13:08:35", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3471.370983483964, "loss": 0.7321006748825312, "entropy": 0.6790678985416889, "sharp": 0.06786358542740345} +{"time": "13:09:33", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3474.949028361901, "loss": 0.6662357449531555, "entropy": 0.6802346408367157, "sharp": 0.05306807532906532} +{"time": "13:10:27", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3471.3741941254825, "loss": 0.6629285104572773, "entropy": 0.6554826572537422, "sharp": 0.11328512243926525} +{"time": "13:11:24", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3476.5469817716526, "loss": 0.6107969712466002, "entropy": 0.6484292279928923, "sharp": 0.13795110397040844} +{"time": "13:12:21", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3479.900163605576, "loss": 0.6779581196606159, "entropy": 0.6427349597215652, "sharp": 0.12921183370053768} +{"time": "13:13:17", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3478.3483786562365, "loss": 0.6582030206918716, "entropy": 0.6541828140616417, "sharp": 0.1097470372915268} +{"time": "13:14:12", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3479.7926639903835, "loss": 0.6291924305260181, "entropy": 0.6428272649645805, "sharp": 0.13084788620471954} +{"time": "13:15:08", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3479.3814033133276, "loss": 0.6468897089362144, "entropy": 0.6616482585668564, "sharp": 0.11564681492745876} +{"time": "13:15:28", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.61387699842453, "noul_T": 0.7502643155888578, "noul_auc": 0.7150427840880323, "noul_acc": 0.662, "noul_f1": 0.7725437415881561, "noul_nll": 0.6118612088561058, "noul_ece": 0.08541951048374176, "noul_brier": 0.21110023911294656, "noul_sharp": 0.1390904188156128, "entropy": 0.6427621841430664} +{"time": "13:16:25", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3372.6053614263637, "loss": 0.6409700810909271, "entropy": 0.6763819307088852, "sharp": 0.07838659267872572} +{"time": "13:17:21", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3385.7325021633624, "loss": 0.6850181668996811, "entropy": 0.6571488790214062, "sharp": 0.11317025031894445} +{"time": "13:18:16", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3390.1002125182627, "loss": 0.6586766093969345, "entropy": 0.6385197266936302, "sharp": 0.1514297528192401} +{"time": "13:19:10", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3397.621575305199, "loss": 0.6709637492895126, "entropy": 0.6612059660255909, "sharp": 0.09924796782433987} +{"time": "13:20:06", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3405.2406784077666, "loss": 0.6333407796919346, "entropy": 0.6513965055346489, "sharp": 0.12322159390896559} +{"time": "13:21:04", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3407.8652339629957, "loss": 0.7106940057128668, "entropy": 0.6496372148394585, "sharp": 0.12053652293980122} +{"time": "13:22:00", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3413.2216141897443, "loss": 0.658356998115778, "entropy": 0.6635720022022724, "sharp": 0.08919562213122845} +{"time": "13:22:57", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3418.242738289302, "loss": 0.6278098672628403, "entropy": 0.6088906601071358, "sharp": 0.17759525030851364} +{"time": "13:23:54", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3421.938308848099, "loss": 0.6532021313905716, "entropy": 0.6763139292597771, "sharp": 0.06488098204135895} +{"time": "13:24:50", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3425.556667401872, "loss": 0.6596585661172867, "entropy": 0.6380285620689392, "sharp": 0.13881313893944025} +{"time": "13:25:10", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5997543931007385, "noul_T": 0.7233739753605402, "noul_auc": 0.7240046581492297, "noul_acc": 0.686, "noul_f1": 0.7773049645390071, "noul_nll": 0.6016065734028816, "noul_ece": 0.06843488967418672, "noul_brier": 0.2064396714760344, "noul_sharp": 0.13926245272159576, "entropy": 0.6347545385360718} +{"time": "13:26:06", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3374.046036556145, "loss": 0.728416483849287, "entropy": 0.6540384255349636, "sharp": 0.12134160567075014} +{"time": "13:27:02", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3379.0642488507056, "loss": 0.6382903531193733, "entropy": 0.6674725078046322, "sharp": 0.08678625710308552} +{"time": "13:27:59", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3384.424142240071, "loss": 0.6751176789402962, "entropy": 0.6498658191412687, "sharp": 0.12566701415926218} +{"time": "13:28:55", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3387.48187026368, "loss": 0.6755280904471874, "entropy": 0.6713389791548252, "sharp": 0.07941348850727081} +{"time": "13:29:52", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3388.626732996153, "loss": 0.6453321930021048, "entropy": 0.6084790658205748, "sharp": 0.18274418823421001} +{"time": "13:30:49", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3393.9630203769734, "loss": 0.6141139790415764, "entropy": 0.6487501412630081, "sharp": 0.11451198626309633} +{"time": "13:31:45", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3396.66036162014, "loss": 0.6606187708675861, "entropy": 0.6173181906342506, "sharp": 0.14164115022867918} +{"time": "13:32:43", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3400.167096216564, "loss": 0.7238562628626823, "entropy": 0.6690664142370224, "sharp": 0.08110342640429735} +{"time": "13:33:41", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3401.623247012768, "loss": 0.6591993570327759, "entropy": 0.6694832891225815, "sharp": 0.08701068628579378} +{"time": "13:34:37", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3403.9309383059826, "loss": 0.6750735864043236, "entropy": 0.633796151727438, "sharp": 0.15186581015586853} +{"time": "13:34:56", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.5924268960952759, "noul_T": 0.8235588341092702, "noul_auc": 0.7304518067205616, "noul_acc": 0.688, "noul_f1": 0.7796610169491526, "noul_nll": 0.5932661187052727, "noul_ece": 0.05689312601089477, "noul_brier": 0.20314560298842282, "noul_sharp": 0.16020257771015167, "entropy": 0.6179916262626648} +{"time": "13:35:54", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3367.2899085251706, "loss": 0.6096322983503342, "entropy": 0.6450892761349678, "sharp": 0.11650849040597677} +{"time": "13:36:51", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3370.4727846061023, "loss": 0.638837231323123, "entropy": 0.6453867815434933, "sharp": 0.1355299586430192} +{"time": "13:37:51", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3374.924883043018, "loss": 0.6462179850786924, "entropy": 0.6524023376405239, "sharp": 0.12154635041952133} +{"time": "13:38:47", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3377.6440212959264, "loss": 0.6600429341197014, "entropy": 0.6612998396158218, "sharp": 0.10717360582202673} +{"time": "13:39:44", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3380.6418262409397, "loss": 0.6605653762817383, "entropy": 0.644973874092102, "sharp": 0.12811066769063473} +{"time": "13:40:42", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3383.0146381997592, "loss": 0.6918214187026024, "entropy": 0.6228439956903458, "sharp": 0.16477921232581139} +{"time": "13:41:39", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3385.06700465791, "loss": 0.637321762740612, "entropy": 0.6542492862790823, "sharp": 0.10681664478033781} +{"time": "13:42:35", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3388.9526831846606, "loss": 0.6609719395637512, "entropy": 0.6759444288909435, "sharp": 0.07788921892642975} +{"time": "13:43:32", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3391.4997841531826, "loss": 0.6586776375770569, "entropy": 0.66489577293396, "sharp": 0.10015012789517641} +{"time": "13:44:27", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3393.92505860903, "loss": 0.6536326557397842, "entropy": 0.6512454897165298, "sharp": 0.1197962062433362} +{"time": "13:44:47", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.598845362663269, "noul_T": 0.735947220148929, "noul_auc": 0.7268400533324331, "noul_acc": 0.704, "noul_f1": 0.765079365079365, "noul_nll": 0.5981243264079094, "noul_ece": 0.08023010075092317, "noul_brier": 0.20508747941090547, "noul_sharp": 0.1429869532585144, "entropy": 0.6325695514678955} +{"time": "13:45:45", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3368.167070312501, "loss": 0.6600133888423443, "entropy": 0.6519585326313972, "sharp": 0.11055010091513395} +{"time": "13:46:41", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3370.061144611227, "loss": 0.6732719950377941, "entropy": 0.6503001637756824, "sharp": 0.11638599820435047} +{"time": "13:47:38", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3372.8565790008774, "loss": 0.6734577342867851, "entropy": 0.6587095707654953, "sharp": 0.10372204333543777} +{"time": "13:48:36", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3374.2355471341493, "loss": 0.6714082881808281, "entropy": 0.6479462385177612, "sharp": 0.12300914525985718} +{"time": "13:49:34", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3377.1354737113575, "loss": 0.6101809777319431, "entropy": 0.6512132808566093, "sharp": 0.11618114821612835} +{"time": "13:50:30", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3378.3794989283783, "loss": 0.6276053637266159, "entropy": 0.6537724919617176, "sharp": 0.11201480776071548} +{"time": "13:51:26", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3380.90028555349, "loss": 0.6744292080402374, "entropy": 0.6492992490530014, "sharp": 0.11929773725569248} +{"time": "13:52:24", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3383.5955451153186, "loss": 0.6026301421225071, "entropy": 0.6428409479558468, "sharp": 0.12657547276467085} +{"time": "13:53:19", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3386.126481119613, "loss": 0.6378263607621193, "entropy": 0.6711188592016697, "sharp": 0.08070253487676382} +{"time": "13:54:14", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3387.689123835015, "loss": 0.6454498060047626, "entropy": 0.6562105901539326, "sharp": 0.10300591960549355} +{"time": "13:54:34", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5951837301254272, "noul_T": 0.8022424860610371, "noul_auc": 0.7221987814551651, "noul_acc": 0.698, "noul_f1": 0.7606973058637084, "noul_nll": 0.5965909399837256, "noul_ece": 0.07016367256641387, "noul_brier": 0.20466233112296456, "noul_sharp": 0.15421432256698608, "entropy": 0.6222149133682251} +{"time": "13:55:16", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9861110, "tok_per_s": 3316.668544463796, "rows_per_s": 5.664608911729549, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5800862908363342, "noul_T": 0.8128870384220848, "choice_ce": 0.668471097946167, "choice_T": 0.9823765467076169, "score_ce": 1.0641282796859741, "score_T": 0.9929303222249871, "noul_auc": 0.7086250666666667, "noul_acc": 0.689, "noul_f1": 0.7663410969196093, "noul_nll": 0.5972189504876733, "noul_ece": 0.041735121816396714, "noul_brier": 0.20542506687095102, "noul_sharp": 0.15873706340789795, "entropy": 0.6186158657073975}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-M1P40-n16k-syn/DONE b/e28/S1-E28-M1P40-n16k-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..e6f0a98e58e95318231fcae5c987ef9902e8901f --- /dev/null +++ b/e28/S1-E28-M1P40-n16k-syn/DONE @@ -0,0 +1 @@ +2026-10-20 15:08:01 \ No newline at end of file diff --git a/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..ead4e0174b2bce3972a9c8fb5891e65e9c1c2470 --- /dev/null +++ b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "out_proj", + "up_proj", + "in_proj_qkv", + "q_proj", + "in_proj_z", + "k_proj", + "v_proj", + "o_proj", + "down_proj", + "gate_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..176258d2405d91b97fca0f6bd942061b79b9e13b --- /dev/null +++ b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7481f94876d4f1a36243fab25f985427ffe69609ed27d53f7a099d6af832b4a3 +size 160486456 diff --git a/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4a958792b891d6bdf66b04fef886504368fec76b --- /dev/null +++ b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0e177b2d6e84ccfe8ccbee745d70cef8bf76f254a4d58eb50a2974d5c28952f +size 176641832 diff --git a/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..b3218e167a44e0cea7678c27ad146637f0f008da --- /dev/null +++ b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.00036888610338792205, + -0.0003691388701554388, + -0.002928646979853511, + -0.0016607323195785284, + -0.0006167158135212958, + 0.00206221966072917, + 0.001397102721966803, + -0.002163229277357459, + -0.0002514572406653315, + -0.0002548548218328506, + -0.00023159636475611478, + 0.0003883895114995539, + -0.0003291864413768053, + -0.0034402175806462765, + -0.0045825704000890255, + -0.00541297672316432, + -0.00575320515781641, + -0.004447372164577246, + -0.004990284331142902, + -0.005246356595307589, + -0.0040365648455917835, + -0.0036311650183051825, + -0.005761807318776846, + -0.00361148570664227 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/calibration.json b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..7064e2d7e326ff5e3cf448e408ca9e20aca27511 --- /dev/null +++ b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.697786954792599, + "choice": 1.0594261278671433, + "score": 0.9790886892108726 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.697786954792599, + "ce_before": 0.5793200731277466, + "ce_after": 0.5707334876060486 + }, + "choice": { + "n": 60, + "T": 1.0594261278671433, + "ce_before": 0.6688334345817566, + "ce_after": 0.6680603623390198 + }, + "score": { + "n": 53, + "T": 0.9790886892108726, + "ce_before": 1.0738638639450073, + "ce_after": 1.0737265348434448 + } + }, + "source": "models/e28/S1-E28-M1P40-n16k-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/head.safetensors b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6d9138e6c423d396503406bd21464df309974c27 --- /dev/null +++ b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:62a35832d7e94b2c43360f78ea68c37cd76b1c273d15ae0c8ef7ab34159a37e8 +size 393464 diff --git a/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/judge_config.json b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..12e9831a30fe2cc19acf5112852da5362530a097 --- /dev/null +++ b/e28/S1-E28-M1P40-n16k-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-M1P40-n16k-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-M1P40-n16k", + "data_manifest": { + "arm": "E28-M1P40-n16k", + "name": null, + "train_sha256": "a5fe7c93585527884a4398047c8086df9a5fa9aca20f19354994863e58ca8e33", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9813191, + "tok_per_s": 3318.4875326110505, + "rows_per_s": 5.695391742419592, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5793200731277466, + "noul_T": 0.697786954792599, + "choice_ce": 0.6688334345817566, + "choice_T": 1.0594261278671433, + "score_ce": 1.0738638639450073, + "score_T": 0.9790886892108726, + "noul_auc": 0.7235456, + "noul_acc": 0.696, + "noul_f1": 0.7734724292101341, + "noul_nll": 0.5940991111397743, + "noul_ece": 0.05802369517087938, + "noul_brier": 0.2031729834036656, + "noul_sharp": 0.1452816277742386, + "entropy": 0.6329353451728821 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-M1P40-n16k-syn/train_log.jsonl b/e28/S1-E28-M1P40-n16k-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..040b770dd98c8440689d1f381c53b78b93d2df4e --- /dev/null +++ b/e28/S1-E28-M1P40-n16k-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "14:18:44", "event": "start", "data": "data/e28/E28-M1P40-n16k", "data_manifest": {"arm": "E28-M1P40-n16k", "name": null, "train_sha256": "a5fe7c93585527884a4398047c8086df9a5fa9aca20f19354994863e58ca8e33", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.73063325881958} +{"time": "14:19:41", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3455.416680614323, "loss": 0.6594873368740082, "entropy": 0.6307111047208309, "sharp": 0.14003448653966188} +{"time": "14:20:36", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3469.18143923776, "loss": 0.6652397066354752, "entropy": 0.6634808331727982, "sharp": 0.11103025637567043} +{"time": "14:21:31", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3465.7124860581293, "loss": 0.7318559214472771, "entropy": 0.6647496744990349, "sharp": 0.07495992071926594} +{"time": "14:22:25", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3458.195484300255, "loss": 0.6385809369385242, "entropy": 0.6579522099345922, "sharp": 0.11998044140636921} +{"time": "14:23:21", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3464.5040740213362, "loss": 0.6582697164267302, "entropy": 0.653070580214262, "sharp": 0.11620954237878323} +{"time": "14:24:18", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3472.2269182927407, "loss": 0.6554521396756172, "entropy": 0.6825159043073654, "sharp": 0.050763838924467564} +{"time": "14:25:14", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3470.3692250054696, "loss": 0.6348754204809666, "entropy": 0.6698445975780487, "sharp": 0.07856910675764084} +{"time": "14:26:09", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3471.0911292913806, "loss": 0.6536184549331665, "entropy": 0.6322706751525402, "sharp": 0.1402834514155984} +{"time": "14:27:06", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3475.416176821334, "loss": 0.66744464635849, "entropy": 0.6745415925979614, "sharp": 0.07394020352512598} +{"time": "14:28:03", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3477.4550838778514, "loss": 0.660002663731575, "entropy": 0.6114483699202538, "sharp": 0.17269742488861084} +{"time": "14:28:23", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.6017172932624817, "noul_T": 0.7527004168507331, "noul_auc": 0.7296754485156368, "noul_acc": 0.708, "noul_f1": 0.7774390243902439, "noul_nll": 0.5988093336820602, "noul_ece": 0.07027411860227586, "noul_brier": 0.20477312079793952, "noul_sharp": 0.14461849629878998, "entropy": 0.633331835269928} +{"time": "14:29:20", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3371.739951645072, "loss": 0.6517752967774868, "entropy": 0.6714483499526978, "sharp": 0.08941776864230633} +{"time": "14:30:14", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3380.1330877297214, "loss": 0.6920490637421608, "entropy": 0.649154856801033, "sharp": 0.12782281078398228} +{"time": "14:31:10", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3387.4819425306723, "loss": 0.6284860000014305, "entropy": 0.665465272963047, "sharp": 0.10028049629181623} +{"time": "14:32:04", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3392.8211795040806, "loss": 0.6713457964360714, "entropy": 0.6687470264732838, "sharp": 0.08854744583368301} +{"time": "14:32:59", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3398.247859306188, "loss": 0.7257119826972485, "entropy": 0.6276344861835241, "sharp": 0.14902007579803467} +{"time": "14:33:55", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3402.0698112429436, "loss": 0.728968545794487, "entropy": 0.6834740117192268, "sharp": 0.0535788768902421} +{"time": "14:34:50", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3406.7246712116294, "loss": 0.7231758609414101, "entropy": 0.6158965528011322, "sharp": 0.18414349295198917} +{"time": "14:35:49", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3411.659169989872, "loss": 0.6231337934732437, "entropy": 0.6587008535861969, "sharp": 0.10693285334855318} +{"time": "14:36:44", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3414.2567577717778, "loss": 0.6817149333655834, "entropy": 0.6542453244328499, "sharp": 0.11074962560087442} +{"time": "14:37:42", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3416.804591656011, "loss": 0.6587735489010811, "entropy": 0.6424405872821808, "sharp": 0.13376750331372023} +{"time": "14:38:02", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5940027832984924, "noul_T": 0.7503689800664315, "noul_auc": 0.7257767801387318, "noul_acc": 0.702, "noul_f1": 0.7682737169517885, "noul_nll": 0.595310007661581, "noul_ece": 0.056828067302703865, "noul_brier": 0.2035034200480485, "noul_sharp": 0.1498262584209442, "entropy": 0.6282942891120911} +{"time": "14:38:58", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3362.6344681890246, "loss": 0.7228566780686378, "entropy": 0.6564287468791008, "sharp": 0.10397548228502274} +{"time": "14:39:54", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3367.2323232936196, "loss": 0.6279414966702461, "entropy": 0.6478557288646698, "sharp": 0.12800051644444466} +{"time": "14:40:52", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3374.4373073199376, "loss": 0.6809984855353832, "entropy": 0.6346354708075523, "sharp": 0.1317371055483818} +{"time": "14:41:48", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3377.8946073617244, "loss": 0.6955315601080656, "entropy": 0.6281479224562645, "sharp": 0.15313439536839724} +{"time": "14:42:47", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3382.8676206029527, "loss": 0.6292684972286224, "entropy": 0.645242802798748, "sharp": 0.1274282019585371} +{"time": "14:43:46", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3386.713740069958, "loss": 0.6442285925149918, "entropy": 0.6384358927607536, "sharp": 0.14295257814228535} +{"time": "14:44:42", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3391.369378766094, "loss": 0.6311197802424431, "entropy": 0.6432098895311356, "sharp": 0.13158029597252607} +{"time": "14:45:40", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3395.6086587602313, "loss": 0.6965556144714355, "entropy": 0.6785937249660492, "sharp": 0.06634129211306572} +{"time": "14:46:35", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3397.797234078489, "loss": 0.6618181969970465, "entropy": 0.6473540551960468, "sharp": 0.13560253474861383} +{"time": "14:47:30", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3400.0758068245796, "loss": 0.6837029457092285, "entropy": 0.6656754948198795, "sharp": 0.09355952125042677} +{"time": "14:47:50", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.6001991033554077, "noul_T": 0.7265663413077567, "noul_auc": 0.7291522505949266, "noul_acc": 0.702, "noul_f1": 0.7682737169517885, "noul_nll": 0.5972377725243568, "noul_ece": 0.0728271441459656, "noul_brier": 0.2040717062719466, "noul_sharp": 0.14454738795757294, "entropy": 0.635818600654602} +{"time": "14:48:46", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3365.7618814763855, "loss": 0.6379623003304005, "entropy": 0.6540741622447968, "sharp": 0.11074624210596085} +{"time": "14:49:41", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3368.925181925026, "loss": 0.6609176695346832, "entropy": 0.6556948348879814, "sharp": 0.11495323106646538} +{"time": "14:50:39", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3372.942536012461, "loss": 0.6416081115603447, "entropy": 0.6643768697977066, "sharp": 0.10546861868351698} +{"time": "14:51:35", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3376.2459151625058, "loss": 0.683039516210556, "entropy": 0.6766052022576332, "sharp": 0.07391216792166233} +{"time": "14:52:32", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3380.8155287467107, "loss": 0.6254632994532585, "entropy": 0.6530685052275658, "sharp": 0.11443551816046238} +{"time": "14:53:28", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3383.887475808015, "loss": 0.6855271868407726, "entropy": 0.6722672451287508, "sharp": 0.07476869598031044} +{"time": "14:54:25", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3386.69496237285, "loss": 0.6844386383891106, "entropy": 0.6680985316634178, "sharp": 0.08075587917119265} +{"time": "14:55:22", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3390.130598235832, "loss": 0.6348525881767273, "entropy": 0.6678259372711182, "sharp": 0.09265239536762238} +{"time": "14:56:19", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3392.673042999973, "loss": 0.6970858424901962, "entropy": 0.6461808048188686, "sharp": 0.13032672181725502} +{"time": "14:57:15", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3395.9836051454777, "loss": 0.6835946682840586, "entropy": 0.6528009697794914, "sharp": 0.11522696260362864} +{"time": "14:57:35", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5936031341552734, "noul_T": 0.6381566797446567, "noul_auc": 0.7378440870196283, "noul_acc": 0.702, "noul_f1": 0.7799113737075333, "noul_nll": 0.5954550780951977, "noul_ece": 0.0767899217605591, "noul_brier": 0.20347306488650185, "noul_sharp": 0.133924201130867, "entropy": 0.6400524973869324} +{"time": "14:58:32", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3370.0535377659207, "loss": 0.6452237907797098, "entropy": 0.6653882581740618, "sharp": 0.09267841931432486} +{"time": "14:59:28", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3371.112824140209, "loss": 0.6597021147608757, "entropy": 0.6682580970227718, "sharp": 0.08280072174966335} +{"time": "15:00:24", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3373.9851996622206, "loss": 0.7026616744697094, "entropy": 0.6504443250596523, "sharp": 0.1117750545963645} +{"time": "15:01:20", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3376.088353423699, "loss": 0.6679081581532955, "entropy": 0.6689461469650269, "sharp": 0.09090832062065601} +{"time": "15:02:17", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3379.106769498312, "loss": 0.6784391924738884, "entropy": 0.6668302714824677, "sharp": 0.09302367828786373} +{"time": "15:03:13", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3381.3771191391506, "loss": 0.6546323038637638, "entropy": 0.650833435356617, "sharp": 0.11893327068537474} +{"time": "15:04:09", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3383.8264350331337, "loss": 0.6913115009665489, "entropy": 0.6531240046024323, "sharp": 0.1155363917350769} +{"time": "15:05:05", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3385.820600626424, "loss": 0.6671162657439709, "entropy": 0.6585468500852585, "sharp": 0.10744495689868927} +{"time": "15:06:01", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3387.6009785604106, "loss": 0.6595820225775242, "entropy": 0.6648876294493675, "sharp": 0.10136905685067177} +{"time": "15:06:59", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3389.9471699044507, "loss": 0.6392969936132431, "entropy": 0.66969945281744, "sharp": 0.09128210600465536} +{"time": "15:07:18", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5946288108825684, "noul_T": 0.6894469533131458, "noul_auc": 0.7333715886651703, "noul_acc": 0.712, "noul_f1": 0.7757009345794392, "noul_nll": 0.5950047305226326, "noul_ece": 0.07031712043285372, "noul_brier": 0.203111133781924, "noul_sharp": 0.14167261123657227, "entropy": 0.6354730129241943} +{"time": "15:08:01", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9813191, "tok_per_s": 3318.4875326110505, "rows_per_s": 5.695391742419592, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5793200731277466, "noul_T": 0.697786954792599, "choice_ce": 0.6688334345817566, "choice_T": 1.0594261278671433, "score_ce": 1.0738638639450073, "score_T": 0.9790886892108726, "noul_auc": 0.7235456, "noul_acc": 0.696, "noul_f1": 0.7734724292101341, "noul_nll": 0.5940991111397743, "noul_ece": 0.05802369517087938, "noul_brier": 0.2031729834036656, "noul_sharp": 0.1452816277742386, "entropy": 0.6329353451728821}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-M3-n16k-syn/DONE b/e28/S1-E28-M3-n16k-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..ae40f191dc432af01abe8f3ae9f4ed50fb9043ae --- /dev/null +++ b/e28/S1-E28-M3-n16k-syn/DONE @@ -0,0 +1 @@ +2026-10-20 09:05:07 \ No newline at end of file diff --git a/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..cbc56e5c281eb373b4e72b9535e0726003c64af5 --- /dev/null +++ b/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "o_proj", + "down_proj", + "q_proj", + "v_proj", + "up_proj", + "k_proj", + "in_proj_z", + "gate_proj", + "out_proj", + "in_proj_qkv" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2b59b5b268deb741ed4e29a1d14bcc4bb9260c0b --- /dev/null +++ b/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a1089d85e292b9021a93142495a8d43b8cba99c267887119b089ed177efc24e +size 160486456 diff --git a/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a1685ceac257c565fa65fbd3729985e72ce6bab3 --- /dev/null +++ b/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e184558c318c9e74c9fe7fdab71dbf876e40088652ab475ebfd1a7ffd3a3d70 +size 176641832 diff --git a/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..5adabbeb8cc0ed4e6649530adc92e79688a496c4 --- /dev/null +++ b/e28/S1-E28-M3-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.00037291122134774923, + -0.0003731649194378406, + -0.002843049820512533, + -0.0019383331527933478, + -0.0005820144433528185, + 0.0018673143349587917, + 0.001045079785399139, + -0.0030142413452267647, + -0.0003019664727617055, + -0.00030960640287958086, + -0.0001781875325832516, + 0.0002645269560161978, + 1.4007212485012133e-05, + -0.0034123838413506746, + -0.0040108924731612206, + -0.0050934976898133755, + -0.005911197047680616, + -0.00434337230399251, + -0.00492717232555151, + -0.004941197112202644, + -0.0037641737144440413, + -0.0033391311299055815, + -0.005663158372044563, + -0.003459150204434991 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-M3-n16k-syn/ckpt-16000/calibration.json b/e28/S1-E28-M3-n16k-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..8c1ca2eb8487a71ff94b31462ab9c664de8dba6e --- /dev/null +++ b/e28/S1-E28-M3-n16k-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.776115087205719, + "choice": 0.9844100756577693, + "score": 0.9613556516074261 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.776115087205719, + "ce_before": 0.5745712518692017, + "ce_after": 0.5699681043624878 + }, + "choice": { + "n": 60, + "T": 0.9844100756577693, + "ce_before": 0.6659258008003235, + "ce_after": 0.6658687591552734 + }, + "score": { + "n": 53, + "T": 0.9613556516074261, + "ce_before": 1.052063226699829, + "ce_after": 1.051601767539978 + } + }, + "source": "models/e28/S1-E28-M3-n16k-syn/ckpt-16000" +} \ No newline at end of file diff --git a/e28/S1-E28-M3-n16k-syn/ckpt-16000/head.safetensors b/e28/S1-E28-M3-n16k-syn/ckpt-16000/head.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4f6af851ade355de35a7048271a2869015ee5409 --- /dev/null +++ b/e28/S1-E28-M3-n16k-syn/ckpt-16000/head.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0dbe8c7022414763115200550638820189d7a56886ab89ee8c5128681edaa71a +size 393464 diff --git a/e28/S1-E28-M3-n16k-syn/ckpt-16000/judge_config.json b/e28/S1-E28-M3-n16k-syn/ckpt-16000/judge_config.json new file mode 100644 index 0000000000000000000000000000000000000000..06364c5fe762f861abbfeecf1245853062dfd0b4 --- /dev/null +++ b/e28/S1-E28-M3-n16k-syn/ckpt-16000/judge_config.json @@ -0,0 +1,164 @@ +{ + "base_model_path": "/root/models/Qwen3.5-9B", + "hidden_size": 4096, + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "kinds": [ + "noul", + "choice", + "score" + ], + "model_name": "jev-judge-qwen35-9b", + "model_version": "0.8.0", + "lm_head_included": true, + "weights_mode": "unmerged", + "adapter_subfolder": "adapter", + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ] + }, + "trained_stage": "e16", + "source_checkpoint": "models/e28/S1-E28-M3-n16k-syn/ckpt-16000", + "softcap": null, + "e16": { + "data": "data/e28/E28-M3-n16k", + "data_manifest": { + "arm": "E28-M3-n16k", + "name": null, + "train_sha256": "7bbfa28cd203e7842b3bba83ff9ce0b6a6068bcd44402d711718195bf9374387", + "n_train": 16000 + }, + "init": "released", + "objective": "s-log", + "sigma": 0.5, + "K": 8, + "reward": "brier", + "clip": 0.2, + "passes": 1, + "lr": 0.0001, + "lr_head": 0.0002, + "rows_per_step": 32, + "train_seed": 0, + "replay": "data/e16/replay/train.jsonl", + "replay_frac": 0.05, + "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", + "max_len": 3072, + "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", + "total_rows": 16000, + "lora": { + "r": 16, + "lora_alpha": 32, + "lora_dropout": 0.05 + }, + "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", + "event": "ckpt", + "tag": "ckpt-16000", + "rows_seen": 16000, + "replay_seen": 842, + "step": 500, + "tokens": 9818661, + "tok_per_s": 3320.17788722394, + "rows_per_s": 5.695118301170042, + "val": { + "n": 1200, + "dropped_long": 0, + "noul_ce": 0.5745712518692017, + "noul_T": 0.776115087205719, + "choice_ce": 0.6659258008003235, + "choice_T": 0.9844100756577693, + "score_ce": 1.052063226699829, + "score_T": 0.9613556516074261, + "noul_auc": 0.7187968, + "noul_acc": 0.691, + "noul_f1": 0.7775377969762419, + "noul_nll": 0.5919548912793398, + "noul_ece": 0.036156127929687504, + "noul_brier": 0.20273978439980542, + "noul_sharp": 0.16080120205879211, + "entropy": 0.6192412376403809 + }, + "dropped_long": 0, + "peak_gib": 18.282634735107422 + } +} \ No newline at end of file diff --git a/e28/S1-E28-M3-n16k-syn/train_log.jsonl b/e28/S1-E28-M3-n16k-syn/train_log.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..5fa3a7ed17030b586672fdb734d8b5ef792652ca --- /dev/null +++ b/e28/S1-E28-M3-n16k-syn/train_log.jsonl @@ -0,0 +1,57 @@ +{"time": "08:15:50", "event": "start", "data": "data/e28/E28-M3-n16k", "data_manifest": {"arm": "E28-M3-n16k", "name": null, "train_sha256": "7bbfa28cd203e7842b3bba83ff9ce0b6a6068bcd44402d711718195bf9374387", "n_train": 16000}, "init": "released", "objective": "s-log", "sigma": 0.5, "K": 8, "reward": "brier", "clip": 0.2, "passes": 1, "lr": 0.0001, "lr_head": 0.0002, "rows_per_step": 32, "train_seed": 0, "replay": "data/e16/replay/train.jsonl", "replay_frac": 0.05, "replay_sha256": "9dd2df19851678750cf79d2fdb47581720930ef2f0c6e18e6fbe23ef2181847e", "max_len": 3072, "schedule": "warmup 0.03 + cosine to x0.05 over 500 steps", "total_rows": 16000, "lora": {"r": 16, "lora_alpha": 32, "lora_dropout": 0.05}, "trainer_sha256": "a0b9d7552588ffb068565686509162a4f122f6f03392e693fa20088493e2a4f3", "device": "cuda", "load_s": 4.75070333480835} +{"time": "08:16:45", "event": "step", "step": 10, "rows_seen": 320, "lr": 7.333333333333333e-05, "tok_per_s": 3435.3679322799203, "loss": 0.6280431635677814, "entropy": 0.653020828962326, "sharp": 0.11968055088073015} +{"time": "08:17:42", "event": "step", "step": 20, "rows_seen": 640, "lr": 9.997508952966197e-05, "tok_per_s": 3457.631104530745, "loss": 0.6748202741146088, "entropy": 0.5950054451823235, "sharp": 0.18847151193767786} +{"time": "08:18:38", "event": "step", "step": 30, "rows_seen": 960, "lr": 9.977596250541417e-05, "tok_per_s": 3472.5677533815187, "loss": 0.7413051947951317, "entropy": 0.661096602678299, "sharp": 0.10585696995258331} +{"time": "08:19:34", "event": "step", "step": 40, "rows_seen": 1280, "lr": 9.937854366485982e-05, "tok_per_s": 3468.6134914082063, "loss": 0.6743080280721188, "entropy": 0.6716899536550045, "sharp": 0.08079732116311789} +{"time": "08:20:29", "event": "step", "step": 50, "rows_seen": 1600, "lr": 9.8784499920729e-05, "tok_per_s": 3468.3489269292104, "loss": 0.6327672637999058, "entropy": 0.6294683292508125, "sharp": 0.14340413082391024} +{"time": "08:21:26", "event": "step", "step": 60, "rows_seen": 1920, "lr": 9.79963228989294e-05, "tok_per_s": 3480.7886580102086, "loss": 0.614691749215126, "entropy": 0.6589956656098366, "sharp": 0.10220550373196602} +{"time": "08:22:21", "event": "step", "step": 70, "rows_seen": 2240, "lr": 9.701731848780134e-05, "tok_per_s": 3479.478410499523, "loss": 0.6867971550673246, "entropy": 0.6158114615827799, "sharp": 0.1618207124993205} +{"time": "08:23:17", "event": "step", "step": 80, "rows_seen": 2560, "lr": 9.585159297207315e-05, "tok_per_s": 3478.192786313143, "loss": 0.699018195271492, "entropy": 0.6602457147091627, "sharp": 0.09780758433043957} +{"time": "08:24:13", "event": "step", "step": 90, "rows_seen": 2880, "lr": 9.450403580967612e-05, "tok_per_s": 3483.3258369135483, "loss": 0.6182956323027611, "entropy": 0.6472971625626087, "sharp": 0.11587055586278439} +{"time": "08:25:10", "event": "step", "step": 100, "rows_seen": 3200, "lr": 9.298029912365855e-05, "tok_per_s": 3486.9580889010394, "loss": 0.6001721117645502, "entropy": 0.661502867937088, "sharp": 0.09925993159413338} +{"time": "08:25:30", "event": "val", "step": 100, "rows_seen": 3200, "n": 500, "dropped_long": 0, "noul_ce": 0.5940777659416199, "noul_T": 0.6373569647379577, "noul_auc": 0.7448988202730755, "noul_acc": 0.704, "noul_f1": 0.7665615141955836, "noul_nll": 0.5918425616621971, "noul_ece": 0.10313586449623109, "noul_brier": 0.20233646139932904, "noul_sharp": 0.13009043037891388, "entropy": 0.6387652158737183} +{"time": "08:26:26", "event": "step", "step": 110, "rows_seen": 3520, "lr": 9.128677399521752e-05, "tok_per_s": 3374.854839763414, "loss": 0.6459744870662689, "entropy": 0.6080095451325178, "sharp": 0.16157883871346712} +{"time": "08:27:21", "event": "step", "step": 120, "rows_seen": 3840, "lr": 8.943056365728335e-05, "tok_per_s": 3384.9653354633747, "loss": 0.6291778646409512, "entropy": 0.6301126666367054, "sharp": 0.16503291111439466} +{"time": "08:28:17", "event": "step", "step": 130, "rows_seen": 4160, "lr": 8.741945370109202e-05, "tok_per_s": 3392.9746326917857, "loss": 0.6678676791489124, "entropy": 0.6823272407054901, "sharp": 0.05957869626581669} +{"time": "08:29:13", "event": "step", "step": 140, "rows_seen": 4480, "lr": 8.526187942070997e-05, "tok_per_s": 3401.261664981343, "loss": 0.6115077547729015, "entropy": 0.6548401862382889, "sharp": 0.1186373820528388} +{"time": "08:30:09", "event": "step", "step": 150, "rows_seen": 4800, "lr": 8.296689043247915e-05, "tok_per_s": 3407.9711070491294, "loss": 0.6777583416551352, "entropy": 0.6230677627027035, "sharp": 0.16279036365449429} +{"time": "08:31:06", "event": "step", "step": 160, "rows_seen": 5120, "lr": 8.054411271778132e-05, "tok_per_s": 3414.5497235510693, "loss": 0.6930452901870012, "entropy": 0.6724376678466797, "sharp": 0.07540644705295563} +{"time": "08:32:01", "event": "step", "step": 170, "rows_seen": 5440, "lr": 7.800370824832674e-05, "tok_per_s": 3418.4821150252433, "loss": 0.6365673951804638, "entropy": 0.6220618858933449, "sharp": 0.17398573644459248} +{"time": "08:32:58", "event": "step", "step": 180, "rows_seen": 5760, "lr": 7.535633236331324e-05, "tok_per_s": 3421.040884069548, "loss": 0.6397799197584391, "entropy": 0.6647312082350254, "sharp": 0.09484176058322191} +{"time": "08:33:55", "event": "step", "step": 190, "rows_seen": 6080, "lr": 7.26130890772308e-05, "tok_per_s": 3425.6752253470268, "loss": 0.6592874564230442, "entropy": 0.5547912418842316, "sharp": 0.22621320839971304} +{"time": "08:34:51", "event": "step", "step": 200, "rows_seen": 6400, "lr": 6.978548450576568e-05, "tok_per_s": 3428.2310036687863, "loss": 0.6437288522720337, "entropy": 0.6706260666251183, "sharp": 0.09018261916935444} +{"time": "08:35:11", "event": "val", "step": 200, "rows_seen": 6400, "n": 500, "dropped_long": 0, "noul_ce": 0.5987415313720703, "noul_T": 0.5709269924858549, "noul_auc": 0.7388398508042058, "noul_acc": 0.708, "noul_f1": 0.7739938080495357, "noul_nll": 0.5996638805866241, "noul_ece": 0.09916960096359255, "noul_brier": 0.20526452563551326, "noul_sharp": 0.1217287927865982, "entropy": 0.6492072343826294} +{"time": "08:36:08", "event": "step", "step": 210, "rows_seen": 6720, "lr": 6.688537860515266e-05, "tok_per_s": 3375.4643857841584, "loss": 0.7253085821866989, "entropy": 0.6469288244843483, "sharp": 0.12065203115344048} +{"time": "08:37:05", "event": "step", "step": 220, "rows_seen": 7040, "lr": 6.392493542739706e-05, "tok_per_s": 3381.4806719724756, "loss": 0.6323906593024731, "entropy": 0.6522977910935879, "sharp": 0.11387774720788002} +{"time": "08:38:00", "event": "step", "step": 230, "rows_seen": 7360, "lr": 6.091657210001309e-05, "tok_per_s": 3385.442378412476, "loss": 0.656912662088871, "entropy": 0.6615087762475014, "sharp": 0.09797994513064623} +{"time": "08:38:55", "event": "step", "step": 240, "rows_seen": 7680, "lr": 5.7872906744276015e-05, "tok_per_s": 3388.9703650754295, "loss": 0.6657134536653757, "entropy": 0.6299861147999763, "sharp": 0.1458584275096655} +{"time": "08:39:50", "event": "step", "step": 250, "rows_seen": 8000, "lr": 5.48067055504363e-05, "tok_per_s": 3392.8793655609516, "loss": 0.6469234228134155, "entropy": 0.6558082178235054, "sharp": 0.10322327259927988} +{"time": "08:40:48", "event": "step", "step": 260, "rows_seen": 8320, "lr": 5.173082923188119e-05, "tok_per_s": 3396.014996637028, "loss": 0.6514052823185921, "entropy": 0.6609690189361572, "sharp": 0.10241662338376045} +{"time": "08:41:44", "event": "step", "step": 270, "rows_seen": 8640, "lr": 4.865817908283331e-05, "tok_per_s": 3399.682753022537, "loss": 0.6968852169811726, "entropy": 0.6440061405301094, "sharp": 0.12291576433926821} +{"time": "08:42:42", "event": "step", "step": 280, "rows_seen": 8960, "lr": 4.560164286583922e-05, "tok_per_s": 3403.303453009007, "loss": 0.6682115644216537, "entropy": 0.6561730057001114, "sharp": 0.11819942947477102} +{"time": "08:43:39", "event": "step", "step": 290, "rows_seen": 9280, "lr": 4.257404075601443e-05, "tok_per_s": 3407.6185602811165, "loss": 0.665389109402895, "entropy": 0.6348862498998642, "sharp": 0.14828883484005928} +{"time": "08:44:34", "event": "step", "step": 300, "rows_seen": 9600, "lr": 3.958807156877371e-05, "tok_per_s": 3410.29770973028, "loss": 0.6619185544550419, "entropy": 0.6530555486679077, "sharp": 0.11806929577142} +{"time": "08:44:54", "event": "val", "step": 300, "rows_seen": 9600, "n": 500, "dropped_long": 0, "noul_ce": 0.6002089977264404, "noul_T": 0.8074674288212367, "noul_auc": 0.7301480143795042, "noul_acc": 0.708, "noul_f1": 0.7645161290322581, "noul_nll": 0.6018208523988724, "noul_ece": 0.07324713563919069, "noul_brier": 0.20605082982167594, "noul_sharp": 0.15163512527942657, "entropy": 0.625432014465332} +{"time": "08:45:51", "event": "step", "step": 310, "rows_seen": 9920, "lr": 3.665625949658603e-05, "tok_per_s": 3374.985699278829, "loss": 0.6582349389791489, "entropy": 0.6504995375871658, "sharp": 0.12240273039788008} +{"time": "08:46:49", "event": "step", "step": 320, "rows_seen": 10240, "lr": 3.379090157815887e-05, "tok_per_s": 3379.533118804338, "loss": 0.6548709832131863, "entropy": 0.6577864475548267, "sharp": 0.1110688615590334} +{"time": "08:47:45", "event": "step", "step": 330, "rows_seen": 10560, "lr": 3.10040161203844e-05, "tok_per_s": 3383.2088012616064, "loss": 0.6516106426715851, "entropy": 0.6369383335113525, "sharp": 0.14506545290350914} +{"time": "08:48:42", "event": "step", "step": 340, "rows_seen": 10880, "lr": 2.8307292289383776e-05, "tok_per_s": 3386.3427739841386, "loss": 0.6735629513859749, "entropy": 0.6184869920834899, "sharp": 0.1602584756910801} +{"time": "08:49:37", "event": "step", "step": 350, "rows_seen": 11200, "lr": 2.571204108208194e-05, "tok_per_s": 3388.872851691382, "loss": 0.6651802025735378, "entropy": 0.6606216467916965, "sharp": 0.10349666140973568} +{"time": "08:50:32", "event": "step", "step": 360, "rows_seen": 11520, "lr": 2.322914788395595e-05, "tok_per_s": 3390.5717734374666, "loss": 0.6615963950753212, "entropy": 0.6453824415802956, "sharp": 0.13271488528698683} +{"time": "08:51:28", "event": "step", "step": 370, "rows_seen": 11840, "lr": 2.0869026811945157e-05, "tok_per_s": 3393.8446146054875, "loss": 0.6210368238389492, "entropy": 0.6663697883486748, "sharp": 0.10000165179371834} +{"time": "08:52:24", "event": "step", "step": 380, "rows_seen": 12160, "lr": 1.8641577034025627e-05, "tok_per_s": 3395.995860957951, "loss": 0.6813357546925545, "entropy": 0.6642145030200481, "sharp": 0.10003217030316591} +{"time": "08:53:21", "event": "step", "step": 390, "rows_seen": 12480, "lr": 1.6556141248659472e-05, "tok_per_s": 3398.0565031387123, "loss": 0.6139695644378662, "entropy": 0.6442504767328501, "sharp": 0.13212024699896574} +{"time": "08:54:19", "event": "step", "step": 400, "rows_seen": 12800, "lr": 1.4621466498270773e-05, "tok_per_s": 3399.998309531958, "loss": 0.6170563511550426, "entropy": 0.6490833908319473, "sharp": 0.12277348060160875} +{"time": "08:54:39", "event": "val", "step": 400, "rows_seen": 12800, "n": 500, "dropped_long": 0, "noul_ce": 0.5909287333488464, "noul_T": 0.7607507364457203, "noul_auc": 0.7317007307893536, "noul_acc": 0.7, "noul_f1": 0.7706422018348624, "noul_nll": 0.5920728217661381, "noul_ece": 0.06440289545059204, "noul_brier": 0.20220617515029904, "noul_sharp": 0.15552270412445068, "entropy": 0.6249164938926697} +{"time": "08:55:36", "event": "step", "step": 410, "rows_seen": 13120, "lr": 1.284566748110963e-05, "tok_per_s": 3374.0735079508045, "loss": 0.6553576439619064, "entropy": 0.6421326175332069, "sharp": 0.13075278233736753} +{"time": "08:56:31", "event": "step", "step": 420, "rows_seen": 13440, "lr": 1.1236192515387821e-05, "tok_per_s": 3375.3614066461255, "loss": 0.6911936439573765, "entropy": 0.6502003520727158, "sharp": 0.12301609572023153} +{"time": "08:57:27", "event": "step", "step": 430, "rows_seen": 13760, "lr": 9.799792298443728e-06, "tok_per_s": 3377.327930834222, "loss": 0.6705067995935678, "entropy": 0.6507012397050858, "sharp": 0.12023940496146679} +{"time": "08:58:26", "event": "step", "step": 440, "rows_seen": 14080, "lr": 8.542491591971554e-06, "tok_per_s": 3380.225530157364, "loss": 0.6368182301521301, "entropy": 0.62137845531106, "sharp": 0.16334370337426662} +{"time": "08:59:23", "event": "step", "step": 450, "rows_seen": 14400, "lr": 7.469563952077155e-06, "tok_per_s": 3382.415183175664, "loss": 0.6034993585199118, "entropy": 0.6488325148820877, "sharp": 0.12000551167875528} +{"time": "09:00:20", "event": "step", "step": 460, "rows_seen": 14720, "lr": 6.585509610150972e-06, "tok_per_s": 3385.5397932457727, "loss": 0.6458639167249203, "entropy": 0.6559086628258228, "sharp": 0.11727746669203043} +{"time": "09:01:16", "event": "step", "step": 470, "rows_seen": 15040, "lr": 5.894036597333646e-06, "tok_per_s": 3387.430180586635, "loss": 0.6495631784200668, "entropy": 0.6433248668909073, "sharp": 0.1303669549524784} +{"time": "09:02:11", "event": "step", "step": 480, "rows_seen": 15360, "lr": 5.398045191744809e-06, "tok_per_s": 3388.9769315050335, "loss": 0.6298693045973778, "entropy": 0.6322994604706764, "sharp": 0.1421762779355049} +{"time": "09:03:07", "event": "step", "step": 490, "rows_seen": 15680, "lr": 5.099615753708697e-06, "tok_per_s": 3390.711445366854, "loss": 0.6378781422972679, "entropy": 0.6281113494187593, "sharp": 0.15428979881107807} +{"time": "09:04:04", "event": "step", "step": 500, "rows_seen": 16000, "lr": 5e-06, "tok_per_s": 3391.7400624538727, "loss": 0.6428715288639069, "entropy": 0.6657595038414001, "sharp": 0.0916857197880745} +{"time": "09:04:24", "event": "val", "step": 500, "rows_seen": 16000, "n": 500, "dropped_long": 0, "noul_ce": 0.5894385576248169, "noul_T": 0.7638104225711005, "noul_auc": 0.7341479468700951, "noul_acc": 0.702, "noul_f1": 0.7772795216741405, "noul_nll": 0.5900010139942169, "noul_ece": 0.05173685491085053, "noul_brier": 0.20130584602782298, "noul_sharp": 0.1556250900030136, "entropy": 0.6230663061141968} +{"time": "09:05:07", "event": "ckpt", "tag": "ckpt-16000", "rows_seen": 16000, "replay_seen": 842, "step": 500, "tokens": 9818661, "tok_per_s": 3320.17788722394, "rows_per_s": 5.695118301170042, "val": {"n": 1200, "dropped_long": 0, "noul_ce": 0.5745712518692017, "noul_T": 0.776115087205719, "choice_ce": 0.6659258008003235, "choice_T": 0.9844100756577693, "score_ce": 1.052063226699829, "score_T": 0.9613556516074261, "noul_auc": 0.7187968, "noul_acc": 0.691, "noul_f1": 0.7775377969762419, "noul_nll": 0.5919548912793398, "noul_ece": 0.036156127929687504, "noul_brier": 0.20273978439980542, "noul_sharp": 0.16080120205879211, "entropy": 0.6192412376403809}, "dropped_long": 0, "peak_gib": 18.282634735107422} diff --git a/e28/S1-E28-MAX-n16k-syn/DONE b/e28/S1-E28-MAX-n16k-syn/DONE new file mode 100644 index 0000000000000000000000000000000000000000..963eb6a80a1d365e6aaa1b8ca70e3975072dec43 --- /dev/null +++ b/e28/S1-E28-MAX-n16k-syn/DONE @@ -0,0 +1 @@ +2026-10-20 18:46:32 \ No newline at end of file diff --git a/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter/README.md b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter/README.md new file mode 100644 index 0000000000000000000000000000000000000000..57167079ef3f181aacdbd678a733d71697d18f63 --- /dev/null +++ b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter/README.md @@ -0,0 +1,206 @@ +--- +base_model: /workspace/jev_kt/models/hf/JEV-9B +library_name: peft +tags: +- base_model:adapter:/workspace/jev_kt/models/hf/JEV-9B +- lora +- transformers +--- + +# Model Card for Model ID + + + + + +## Model Details + +### Model Description + + + + + +- **Developed by:** [More Information Needed] +- **Funded by [optional]:** [More Information Needed] +- **Shared by [optional]:** [More Information Needed] +- **Model type:** [More Information Needed] +- **Language(s) (NLP):** [More Information Needed] +- **License:** [More Information Needed] +- **Finetuned from model [optional]:** [More Information Needed] + +### Model Sources [optional] + + + +- **Repository:** [More Information Needed] +- **Paper [optional]:** [More Information Needed] +- **Demo [optional]:** [More Information Needed] + +## Uses + + + +### Direct Use + + + +[More Information Needed] + +### Downstream Use [optional] + + + +[More Information Needed] + +### Out-of-Scope Use + + + +[More Information Needed] + +## Bias, Risks, and Limitations + + + +[More Information Needed] + +### Recommendations + + + +Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. + +## How to Get Started with the Model + +Use the code below to get started with the model. + +[More Information Needed] + +## Training Details + +### Training Data + + + +[More Information Needed] + +### Training Procedure + + + +#### Preprocessing [optional] + +[More Information Needed] + + +#### Training Hyperparameters + +- **Training regime:** [More Information Needed] + +#### Speeds, Sizes, Times [optional] + + + +[More Information Needed] + +## Evaluation + + + +### Testing Data, Factors & Metrics + +#### Testing Data + + + +[More Information Needed] + +#### Factors + + + +[More Information Needed] + +#### Metrics + + + +[More Information Needed] + +### Results + +[More Information Needed] + +#### Summary + + + +## Model Examination [optional] + + + +[More Information Needed] + +## Environmental Impact + + + +Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). + +- **Hardware Type:** [More Information Needed] +- **Hours used:** [More Information Needed] +- **Cloud Provider:** [More Information Needed] +- **Compute Region:** [More Information Needed] +- **Carbon Emitted:** [More Information Needed] + +## Technical Specifications [optional] + +### Model Architecture and Objective + +[More Information Needed] + +### Compute Infrastructure + +[More Information Needed] + +#### Hardware + +[More Information Needed] + +#### Software + +[More Information Needed] + +## Citation [optional] + + + +**BibTeX:** + +[More Information Needed] + +**APA:** + +[More Information Needed] + +## Glossary [optional] + + + +[More Information Needed] + +## More Information [optional] + +[More Information Needed] + +## Model Card Authors [optional] + +[More Information Needed] + +## Model Card Contact + +[More Information Needed] +### Framework versions + +- PEFT 0.21.2 \ No newline at end of file diff --git a/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter/adapter_config.json b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..8a4fd029111b685ca1579afb5a09394de80a1ee4 --- /dev/null +++ b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter/adapter_config.json @@ -0,0 +1,57 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": "/workspace/jev_kt/models/hf/JEV-9B", + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 32, + "lora_bias": false, + "lora_dropout": 0.05, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 16, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "up_proj", + "out_proj", + "q_proj", + "gate_proj", + "o_proj", + "k_proj", + "in_proj_z", + "v_proj", + "in_proj_qkv" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4fc18cd1ccda1fff5ed6337a92841c60059e71a8 --- /dev/null +++ b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:23be1a3640863cb9ef7e65df07bda6eccc70850383354c2007509a7001468082 +size 160486456 diff --git a/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..2d093d03334d1bcd8511a3867adb084f382fd3c5 --- /dev/null +++ b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter_vllm/adapter_config.json @@ -0,0 +1,58 @@ +{ + "alora_invocation_tokens": null, + "alpha_pattern": {}, + "arrow_config": null, + "auto_mapping": { + "base_model_class": "Qwen3_5ForCausalLM", + "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5" + }, + "base_model_name_or_path": null, + "bias": "none", + "corda_config": null, + "ensure_weight_tying": false, + "eva_config": null, + "exclude_modules": null, + "fan_in_fan_out": false, + "inference_mode": true, + "init_lora_weights": true, + "kasa_config": null, + "layer_replication": null, + "layers_pattern": null, + "layers_to_transform": null, + "loftq_config": {}, + "lora_alpha": 64.0, + "lora_bias": false, + "lora_dropout": 0.0, + "lora_ga_config": null, + "megatron_config": null, + "megatron_core": "megatron.core", + "modules_to_save": null, + "monteclora_config": null, + "peft_type": "LORA", + "peft_version": "0.21.0", + "qalora_group_size": 16, + "r": 32, + "rank_pattern": {}, + "revision": null, + "target_modules": [ + "down_proj", + "gate_proj", + "in_proj_qkv", + "in_proj_z", + "k_proj", + "lm_head", + "o_proj", + "out_proj", + "q_proj", + "up_proj", + "v_proj" + ], + "target_parameters": null, + "task_type": null, + "trainable_token_indices": null, + "use_bdlora": null, + "use_dora": false, + "use_qalora": false, + "use_rslora": false, + "velora_config": null +} \ No newline at end of file diff --git a/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0c0f3205b40e2e01f29b58f8e7a593bf4116c1ca --- /dev/null +++ b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter_vllm/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9bca96f774f7bcbea07def86933c124d5699da1f00255102ecb948355ffc6a89 +size 176641832 diff --git a/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json new file mode 100644 index 0000000000000000000000000000000000000000..3f64bddb4dce88fa7e9f00fccd484b7003cba9b8 --- /dev/null +++ b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/adapter_vllm/decision_head.json @@ -0,0 +1,99 @@ +{ + "bias": [ + 0.0004556314670480788, + -0.00045588522334583104, + -0.002236202359199524, + -0.0019057613098993897, + -0.0005231379764154553, + 0.001843838719651103, + 0.0010458272881805897, + -0.002899766666814685, + -0.0008054422796703875, + -0.0006466050399467349, + -0.00035108099109493196, + 0.0001508592104073614, + -0.0008660060120746493, + -0.0034646238200366497, + -0.004028933588415384, + -0.005241736304014921, + -0.005826023872941732, + -0.004408507142215967, + -0.004949841648340225, + -0.0050104474648833275, + -0.003780160564929247, + -0.0034287977032363415, + -0.005703889299184084, + -0.0035114421043545008 + ], + "verbalizer_ids": [ + 3721, + 1802, + 15, + 16, + 17, + 18, + 19, + 20, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47 + ], + "slots": { + "num_slots": 24, + "ranges": { + "noul": [ + 0, + 2 + ], + "score": [ + 2, + 8 + ], + "choice": [ + 8, + 24 + ] + }, + "verbalizers": [ + "false", + "true", + "0", + "1", + "2", + "3", + "4", + "5", + "A", + "B", + "C", + "D", + "E", + "F", + "G", + "H", + "I", + "J", + "K", + "L", + "M", + "N", + "O", + "P" + ], + "template_version": "bare-v1" + }, + "note": "decision logits = lm_head-LoRA logprobs of verbalizer ids + bias, then per-kind temperature" +} \ No newline at end of file diff --git a/e28/S1-E28-MAX-n16k-syn/ckpt-16000/calibration.json b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..fe9150e7d4c6d9d40a4b5f1113ef94e01c241eb9 --- /dev/null +++ b/e28/S1-E28-MAX-n16k-syn/ckpt-16000/calibration.json @@ -0,0 +1,29 @@ +{ + "version": 1, + "per_kind": { + "noul": 0.8826742451464699, + "choice": 1.070939002259726, + "score": 0.9724651878555775 + }, + "fit": { + "noul": { + "n": 1087, + "T": 0.8826742451464699, + "ce_before": 0.5732895135879517, + "ce_after": 0.5720655918121338 + }, + "choice": { + "n": 60, + "T": 1.070939002259726, + "ce_before": 0.6842921376228333, + "ce_after": 0.6831218600273132 + }, + "score": { + "n": 53, + "T": 0.9724651878555775, + "ce_before": 1.0540225505828857, + "ce_after": 1.053795576095581 + } + }, + "source": "models/e28/S1-E28-MAX-n16k-syn/ckpt-16000" +} \ No newline at end of file