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": 76725.70653437298,
-      "required_pages_per_second_upper_estimate": 0.005308062406214577
+      "required_token_positions_per_second": 77046.35298508966,
+      "required_pages_per_second_upper_estimate": 0.005330245471677426
     },
     {
       "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": 9590.713316796622,
-      "required_pages_per_second_upper_estimate": 0.004015431388874449
+      "required_token_positions_per_second": 9630.794123136207,
+      "required_pages_per_second_upper_estimate": 0.0040322123855064205
     },
     {
       "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": 19181.426633593244,
-      "required_pages_per_second_upper_estimate": 0.003994985319540324
+      "required_token_positions_per_second": 19261.588246272415,
+      "required_pages_per_second_upper_estimate": 0.004011680869457509
     },
     {
       "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": 28772.13995038987,
-      "required_pages_per_second_upper_estimate": 0.0017248458575119434
+      "required_token_positions_per_second": 28892.382369408624,
+      "required_pages_per_second_upper_estimate": 0.0017320542069325748
     },
     {
       "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": 28772.13995038987,
-      "required_pages_per_second_upper_estimate": 0.0022369073258739435
+      "required_token_positions_per_second": 28892.382369408624,
+      "required_pages_per_second_upper_estimate": 0.0022462556450621456
     },
     {
       "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": 28772.13995038987,
-      "required_pages_per_second_upper_estimate": 0.36905811816664574
+      "required_token_positions_per_second": 28892.382369408624,
+      "required_pages_per_second_upper_estimate": 0.3706004588115676
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06663089064056749,
-    "producer_p95_seconds": 0.17399142399972334,
-    "gpu_update_seconds": 0.9609295675494365,
-    "producer_batches_per_second": 15.008053927935356,
-    "gpu_batches_per_second": 1.0406589970482447,
-    "producer_headroom": 14.421682770729543,
+    "producer_batch_seconds": 0.4727236398984189,
+    "producer_p95_seconds": 0.182360816000255,
+    "gpu_update_seconds": 0.9569304339981953,
+    "producer_batches_per_second": 2.115400871881264,
+    "gpu_batches_per_second": 1.0450080428750226,
+    "producer_headroom": 2.0242914744094986,
     "sustainable": true,
-    "buffer_seconds": 60.5385627556145,
+    "buffer_seconds": 60.286617341886306,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382612.47776408773,
-    "upload_bytes_per_second": 5859858.298880527,
+    "generation_bytes_per_second": 352580.33950493287,
+    "upload_bytes_per_second": 5861222.275304534,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.09973096553777,
+    "estimated_full_shards_per_hour": 75.65553320752134,
     "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": [
@@ -364,6 +364,12 @@
       "bytes": 1076403,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.025973077999879,
+      "bytes": 6061719,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -42,3 +42,4 @@
 {"step": 489200, "loss": 0.5819711685180664, "lr": 0.0003, "grad_norm": 0.17574398380406173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71602755677, "tokens_per_second": 151072.58632100123, "data_wait_s": 1.4813999769103248e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 489300, "loss": 0.6015153527259827, "lr": 0.0003, "grad_norm": 0.16216718847017506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71617227678, "tokens_per_second": 150889.70193210422, "data_wait_s": 7.884999831730966e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 489400, "loss": 0.5750545263290405, "lr": 0.0003, "grad_norm": 0.15705917563219185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71631742681, "tokens_per_second": 134105.13164224813, "data_wait_s": 7.491000360460021e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 489500, "loss": 0.6159577369689941, "lr": 0.0003, "grad_norm": 0.16308084540767134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71646345379, "tokens_per_second": 152161.29341248362, "data_wait_s": 1.3234999642008916e-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:b7d3cdc9be7e2dd094f3f6b0f0b4d97d09223b0f494c867eb4c466a9fdaf575d
-size 93357
+oid sha256:6a447c94f29aa81a2f675ab452241ab9ed1010b06912c7149b82951873f9b8fe
+size 95475