Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75441.74424833333,
-      "required_pages_per_second_upper_estimate": 0.00653909918609282
+      "required_token_positions_per_second": 75438.26811864982,
+      "required_pages_per_second_upper_estimate": 0.006538797883982024
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9430.218031041666,
-      "required_pages_per_second_upper_estimate": 0.0037618549669066804
+      "required_token_positions_per_second": 9429.783514831228,
+      "required_pages_per_second_upper_estimate": 0.0037616816318937402
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18860.436062083332,
-      "required_pages_per_second_upper_estimate": 0.004363029950625184
+      "required_token_positions_per_second": 18859.567029662456,
+      "required_pages_per_second_upper_estimate": 0.004362828915269062
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28290.654093125,
-      "required_pages_per_second_upper_estimate": 0.001722752404680529
+      "required_token_positions_per_second": 28289.35054449368,
+      "required_pages_per_second_upper_estimate": 0.0017226730254080729
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28290.654093125,
-      "required_pages_per_second_upper_estimate": 0.002177892865631582
+      "required_token_positions_per_second": 28289.35054449368,
+      "required_pages_per_second_upper_estimate": 0.0021777925148494894
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28290.654093125,
-      "required_pages_per_second_upper_estimate": 0.3512752411081242
+      "required_token_positions_per_second": 28289.35054449368,
+      "required_pages_per_second_upper_estimate": 0.3512590553968199
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3165663093672606,
-    "producer_p95_seconds": 0.28980844100078684,
-    "gpu_update_seconds": 0.9772838729352259,
-    "producer_batches_per_second": 3.158895847125229,
-    "gpu_batches_per_second": 1.0232441439932363,
-    "producer_headroom": 3.087137967677545,
+    "producer_batch_seconds": 0.3124060307421672,
+    "producer_p95_seconds": 0.2894544330010831,
+    "gpu_update_seconds": 0.9773289053248161,
+    "producer_batches_per_second": 3.200962534635937,
+    "gpu_batches_per_second": 1.02319699596693,
+    "producer_headroom": 3.128393209961489,
     "sustainable": true,
-    "buffer_seconds": 60.59160012198401,
+    "buffer_seconds": 60.594392130138594,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 414353.756387938,
-    "upload_bytes_per_second": 5259383.325062393,
+    "generation_bytes_per_second": 411655.4485560042,
+    "upload_bytes_per_second": 5276415.064103985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.91067045906644,
+    "estimated_full_shards_per_hour": 88.331676411725,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.226145381000606,
-      "bytes": 7407097,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.7928119780008274,
       "bytes": 1141631,
@@ -562,6 +556,12 @@
       "bytes": 9868032,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.1347405200012872,
+      "bytes": 8295093,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -152,3 +152,4 @@
 {"step": 170200, "loss": 0.5940169133245945, "lr": 0.0003, "grad_norm": 0.1530880732622695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24909840087, "tokens_per_second": 151088.30017296015, "data_wait_s": 1.4093000572756864e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 170300, "loss": 0.5909730903804302, "lr": 0.0003, "grad_norm": 0.14221754957390548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24924519187, "tokens_per_second": 149268.1498177246, "data_wait_s": 1.3880000551580451e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 170400, "loss": 0.6160340644419193, "lr": 0.0003, "grad_norm": 0.14022674271114938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24939057822, "tokens_per_second": 148674.73513185445, "data_wait_s": 1.947599957929924e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 170500, "loss": 0.6315521895885468, "lr": 0.0003, "grad_norm": 0.13683199122330644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24953585772, "tokens_per_second": 148597.2368219054, "data_wait_s": 9.371000487590209e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cf88d6998489f60de3d6851967e517f8f3426eaa349783a33e506f458965a26c
-size 328686
+oid sha256:cd71522cfff1570a50df608543f458ef518b9e1ae280292d5ab59c803d5974fa
+size 330804