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": 75719.53877602016,
-      "required_pages_per_second_upper_estimate": 0.00523845338606088
+      "required_token_positions_per_second": 74614.70148775534,
+      "required_pages_per_second_upper_estimate": 0.00516201817888302
     },
     {
       "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": 9464.94234700252,
-      "required_pages_per_second_upper_estimate": 0.003962773710218165
+      "required_token_positions_per_second": 9326.837685969418,
+      "required_pages_per_second_upper_estimate": 0.003904952172596873
     },
     {
       "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": 18929.88469400504,
-      "required_pages_per_second_upper_estimate": 0.003942595767131139
+      "required_token_positions_per_second": 18653.675371938836,
+      "required_pages_per_second_upper_estimate": 0.003885068649474408
     },
     {
       "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": 28394.827041007557,
-      "required_pages_per_second_upper_estimate": 0.0017022265247179278
+      "required_token_positions_per_second": 27980.513057908254,
+      "required_pages_per_second_upper_estimate": 0.0016773890340519469
     },
     {
       "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": 28394.827041007557,
-      "required_pages_per_second_upper_estimate": 0.0022075729067935626
+      "required_token_positions_per_second": 27980.513057908254,
+      "required_pages_per_second_upper_estimate": 0.002175361816982209
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28394.827041007557,
-      "required_pages_per_second_upper_estimate": 0.36421835329212754
+      "required_token_positions_per_second": 27980.513057908254,
+      "required_pages_per_second_upper_estimate": 0.35890397837262544
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11062534689062886,
-    "producer_p95_seconds": 0.27403371699983836,
-    "gpu_update_seconds": 0.9736984824760864,
-    "producer_batches_per_second": 9.039519677065172,
-    "gpu_batches_per_second": 1.0270119734160719,
-    "producer_headroom": 8.80176659187108,
+    "producer_batch_seconds": 0.1279576056016083,
+    "producer_p95_seconds": 0.2960812830024224,
+    "gpu_update_seconds": 0.9881162630141882,
+    "producer_batches_per_second": 7.815088406026184,
+    "gpu_batches_per_second": 1.0120266586338345,
+    "producer_headroom": 7.722215950888101,
     "sustainable": true,
-    "buffer_seconds": 60.36930591351735,
+    "buffer_seconds": 60.27509204386548,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344110.17138441326,
-    "upload_bytes_per_second": 5293674.472625975,
+    "generation_bytes_per_second": 385949.0338716134,
+    "upload_bytes_per_second": 5287439.813373273,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.83803230428026,
+    "estimated_full_shards_per_hour": 82.8156782351618,
     "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": 0.9700920809991658,
-      "bytes": 6347868,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.00576553800056,
       "bytes": 8295904,
@@ -562,6 +556,12 @@
       "bytes": 6126779,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.3350182510002924,
+      "bytes": 7769901,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -231,3 +231,4 @@
 {"step": 627500, "loss": 0.5221641063690186, "lr": 0.0001, "grad_norm": 0.15815645773418807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91849452402, "tokens_per_second": 152911.36024273763, "data_wait_s": 1.022600190481171e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 627600, "loss": 0.5134794116020203, "lr": 0.0001, "grad_norm": 0.1619216194800924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91864129114, "tokens_per_second": 152124.95722376273, "data_wait_s": 1.1240997991990298e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 627700, "loss": 0.5677657127380371, "lr": 0.0001, "grad_norm": 0.16549380138435527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91878657875, "tokens_per_second": 150825.43883612132, "data_wait_s": 1.5724002878414467e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 627800, "loss": 0.5317293405532837, "lr": 0.0001, "grad_norm": 0.17683979314352088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91893147045, "tokens_per_second": 149404.02428802187, "data_wait_s": 1.501199949416332e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:863378fd8a6fc9889da9f8c963254afb246f617b0cf965575f1ddc23eea8a979
-size 497574
+oid sha256:a4685af63efb32fce66cd8dbbcd018cafc5c72af9fdea4ee8cf18b6f87886989
+size 499692