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": 77388.76318187494,
-      "required_pages_per_second_upper_estimate": 0.0053539342035921495
+      "required_token_positions_per_second": 77228.74647643453,
+      "required_pages_per_second_upper_estimate": 0.005342863876619872
     },
     {
       "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": 9673.595397734367,
-      "required_pages_per_second_upper_estimate": 0.00405013238538842
+      "required_token_positions_per_second": 9653.593309554317,
+      "required_pages_per_second_upper_estimate": 0.004041757928758531
     },
     {
       "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": 19347.190795468734,
-      "required_pages_per_second_upper_estimate": 0.00402950962296407
+      "required_token_positions_per_second": 19307.186619108634,
+      "required_pages_per_second_upper_estimate": 0.0040211778080093364
     },
     {
       "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": 29020.786193203105,
-      "required_pages_per_second_upper_estimate": 0.0017397518201077672
+      "required_token_positions_per_second": 28960.779928662952,
+      "required_pages_per_second_upper_estimate": 0.0017361545361728459
     },
     {
       "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": 29020.786193203105,
-      "required_pages_per_second_upper_estimate": 0.002256238477573434
+      "required_token_positions_per_second": 28960.779928662952,
+      "required_pages_per_second_upper_estimate": 0.002251573254445101
     },
     {
       "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": 29020.786193203105,
-      "required_pages_per_second_upper_estimate": 0.3722474851939188
+      "required_token_positions_per_second": 28960.779928662952,
+      "required_pages_per_second_upper_estimate": 0.3714777892621048
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11156282515605653,
-    "producer_p95_seconds": 0.30862537599750794,
-    "gpu_update_seconds": 0.9526964506039254,
-    "producer_batches_per_second": 8.963559309305568,
-    "gpu_batches_per_second": 1.0496522783999964,
-    "producer_headroom": 8.539551138753188,
+    "producer_batch_seconds": 0.13543944014855924,
+    "producer_p95_seconds": 0.32891612499952316,
+    "gpu_update_seconds": 0.9546704221399896,
+    "producer_batches_per_second": 7.383373697522166,
+    "gpu_batches_per_second": 1.047481912929071,
+    "producer_headroom": 7.048688484630783,
     "sustainable": true,
-    "buffer_seconds": 60.0198763880473,
+    "buffer_seconds": 60.14423659481935,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 535823.756870315,
-    "upload_bytes_per_second": 5961717.847362944,
+    "generation_bytes_per_second": 554538.3899737052,
+    "upload_bytes_per_second": 5936185.662543877,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.97530488569342,
+    "estimated_full_shards_per_hour": 118.99102949531904,
     "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": 1.0610679239980527,
-      "bytes": 5725336,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9264001560004544,
       "bytes": 6158826,
@@ -562,6 +556,12 @@
       "bytes": 5956252,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8562311600035173,
+      "bytes": 2647079,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -315,3 +315,4 @@
 {"step": 516500, "loss": 0.5942535400390625, "lr": 0.0003, "grad_norm": 0.16532523981230376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75597973591, "tokens_per_second": 153758.60466063625, "data_wait_s": 2.15519976336509e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 516600, "loss": 0.5852409601211548, "lr": 0.0003, "grad_norm": 0.16251029845161472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75612706432, "tokens_per_second": 153456.53370417174, "data_wait_s": 1.7587997717782855e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 516700, "loss": 0.5686122179031372, "lr": 0.0003, "grad_norm": 0.16759162474548717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75627365397, "tokens_per_second": 152800.1211876946, "data_wait_s": 1.776799763320014e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 516800, "loss": 0.5631062984466553, "lr": 0.0003, "grad_norm": 0.155585051464483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75641993544, "tokens_per_second": 152525.13666541467, "data_wait_s": 1.7769001715350896e-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:fd5b149883b051789a25687f2eef953fb8e002f64da72e4efc289b98765e21f0
-size 676269
+oid sha256:494c18e76cf37661e06c3b48178faa21e7309a230ca5e0997b66fe3670784d99
+size 678387