Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77091.01706168175,
-      "required_pages_per_second_upper_estimate": 0.005333335436131003
+      "required_token_positions_per_second": 77481.69297491993,
+      "required_pages_per_second_upper_estimate": 0.005360363302301829
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9636.377132710219,
-      "required_pages_per_second_upper_estimate": 0.004034549875028562
+      "required_token_positions_per_second": 9685.211621864992,
+      "required_pages_per_second_upper_estimate": 0.0040549958558575685
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19272.754265420437,
-      "required_pages_per_second_upper_estimate": 0.0040140064567783145
+      "required_token_positions_per_second": 19370.423243729983,
+      "required_pages_per_second_upper_estimate": 0.004034348329256026
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28909.13139813066,
-      "required_pages_per_second_upper_estimate": 0.0017330582856301747
+      "required_token_positions_per_second": 29055.634865594977,
+      "required_pages_per_second_upper_estimate": 0.0017418409448068182
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28909.13139813066,
-      "required_pages_per_second_upper_estimate": 0.0022475578083740923
+      "required_token_positions_per_second": 29055.634865594977,
+      "required_pages_per_second_upper_estimate": 0.0022589478085687916
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28909.13139813066,
-      "required_pages_per_second_upper_estimate": 0.370815297368308
+      "required_token_positions_per_second": 29055.634865594977,
+      "required_pages_per_second_upper_estimate": 0.3726944865457726
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07220532289063897,
-    "producer_p95_seconds": 0.19208744400020805,
-    "gpu_update_seconds": 0.956376019024487,
-    "producer_batches_per_second": 13.849394476285136,
-    "gpu_batches_per_second": 1.0456138381847027,
-    "producer_headroom": 13.245228755129299,
+    "producer_batch_seconds": 0.1029256561562022,
+    "producer_p95_seconds": 0.20403107999845815,
+    "gpu_update_seconds": 0.9515538079926187,
+    "producer_batches_per_second": 9.715750546028858,
+    "gpu_batches_per_second": 1.0509127193863923,
+    "producer_headroom": 9.245059429580124,
     "sustainable": true,
-    "buffer_seconds": 60.25168919854268,
+    "buffer_seconds": 60.899443711527596,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 234800.8442365872,
-    "upload_bytes_per_second": 5903167.997206218,
+    "generation_bytes_per_second": 224933.8362153045,
+    "upload_bytes_per_second": 5898419.323375425,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.38279528926098,
+    "estimated_full_shards_per_hour": 48.26556506008483,
     "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": [
@@ -526,6 +526,12 @@
       "bytes": 15672429,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.1264393319997907,
+      "bytes": 6331952,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -74,3 +74,4 @@
 {"step": 492400, "loss": 0.5546084642410278, "lr": 0.0003, "grad_norm": 0.14702425554910908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72070868488, "tokens_per_second": 145653.51486198275, "data_wait_s": 1.738399987516459e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 492500, "loss": 0.5923275351524353, "lr": 0.0003, "grad_norm": 0.14910898405207795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72085369174, "tokens_per_second": 151148.39238931972, "data_wait_s": 1.1112999345641583e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 492600, "loss": 0.5887793302536011, "lr": 0.0003, "grad_norm": 0.15616353349922218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72100060969, "tokens_per_second": 153020.56320353612, "data_wait_s": 1.0361998647567816e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 492700, "loss": 0.5844504833221436, "lr": 0.0003, "grad_norm": 0.1607671126032676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72114861262, "tokens_per_second": 154501.30251059355, "data_wait_s": 1.7562000721227378e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:117ab70e24978815691c165eb1d41677df06f9700f7fc37d83de41b643ef1d32
-size 161916
+oid sha256:0e502af45b381b1a85bde65e1b217050138ac9ab19a7d095dce08b6a29100951
+size 164034