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": 76954.50835702986,
-      "required_pages_per_second_upper_estimate": 0.005323891447199351
+      "required_token_positions_per_second": 76550.8044390162,
+      "required_pages_per_second_upper_estimate": 0.005295962273429021
     },
     {
       "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": 9619.313544628732,
-      "required_pages_per_second_upper_estimate": 0.004027405707027082
+      "required_token_positions_per_second": 9568.850554877024,
+      "required_pages_per_second_upper_estimate": 0.004006277907005098
     },
     {
       "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": 19238.627089257465,
-      "required_pages_per_second_upper_estimate": 0.004006898665977725
+      "required_token_positions_per_second": 19137.70110975405,
+      "required_pages_per_second_upper_estimate": 0.003985878446044228
     },
     {
       "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": 28857.9406338862,
-      "required_pages_per_second_upper_estimate": 0.0017299894774774904
+      "required_token_positions_per_second": 28706.551664631075,
+      "required_pages_per_second_upper_estimate": 0.0017209139399282174
     },
     {
       "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": 28857.9406338862,
-      "required_pages_per_second_upper_estimate": 0.002243577951618461
+      "required_token_positions_per_second": 28706.551664631075,
+      "required_pages_per_second_upper_estimate": 0.0022318081251485778
     },
     {
       "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": 28857.9406338862,
-      "required_pages_per_second_upper_estimate": 0.37015867720894036
+      "required_token_positions_per_second": 28706.551664631075,
+      "required_pages_per_second_upper_estimate": 0.36821682206014644
     }
   ],
   "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": 0.07867285049997008,
-    "producer_p95_seconds": 0.275796330999583,
-    "gpu_update_seconds": 0.9580725232879079,
-    "producer_batches_per_second": 12.710865230444146,
-    "gpu_batches_per_second": 1.0437623203807218,
-    "producer_headroom": 12.177930724504158,
+    "producer_batch_seconds": 0.13560726082016572,
+    "producer_p95_seconds": 0.3182827320015349,
+    "gpu_update_seconds": 0.9631250845800716,
+    "producer_batches_per_second": 7.374236408521964,
+    "gpu_batches_per_second": 1.0382867355552328,
+    "producer_headroom": 7.1023120646711595,
     "sustainable": true,
-    "buffer_seconds": 60.3585689671382,
+    "buffer_seconds": 60.67688032854451,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 454040.5014910162,
-    "upload_bytes_per_second": 6117747.92120753,
+    "generation_bytes_per_second": 451694.6317912985,
+    "upload_bytes_per_second": 6090694.188556159,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.42652209804406,
+    "estimated_full_shards_per_hour": 96.92315306953637,
     "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": 1.3789669100005995,
-      "bytes": 6252708,
-      "pages": 25,
-      "shards": 3
-    },
     {
       "seconds": 1.107486043998506,
       "bytes": 11895475,
@@ -562,6 +556,12 @@
       "bytes": 9995304,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.8846473650010012,
+      "bytes": 1315669,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -295,3 +295,4 @@
 {"step": 514500, "loss": 0.5570145845413208, "lr": 0.0003, "grad_norm": 0.1675595931676757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75305303347, "tokens_per_second": 152018.73569825958, "data_wait_s": 1.6735997633077204e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 514600, "loss": 0.6139761209487915, "lr": 0.0003, "grad_norm": 0.1643211878055086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75320020613, "tokens_per_second": 153367.62589523522, "data_wait_s": 1.5787001757416874e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 514700, "loss": 0.6013380885124207, "lr": 0.0003, "grad_norm": 0.15445432625167307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75334633324, "tokens_per_second": 151803.7379134772, "data_wait_s": 1.4360000932356343e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 514800, "loss": 0.5757641196250916, "lr": 0.0003, "grad_norm": 0.18948936053082832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75349244454, "tokens_per_second": 152224.3600448378, "data_wait_s": 1.2065000191796571e-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:dbcc6ace69610880cfd55358639aa24fb06db3118d87eab844e95397938e11ea
-size 633126
+oid sha256:4869186dba3b17f6f1e786538f9cd38a0a38d259919d781c0e5720e890f22057
+size 635244