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": 76137.40519766882,
-      "required_pages_per_second_upper_estimate": 0.005267362354694215
+      "required_token_positions_per_second": 76947.63837552589,
+      "required_pages_per_second_upper_estimate": 0.005323416165938353
     },
     {
       "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": 9517.175649708603,
-      "required_pages_per_second_upper_estimate": 0.003984642703305808
+      "required_token_positions_per_second": 9618.454796940736,
+      "required_pages_per_second_upper_estimate": 0.004027046167302809
     },
     {
       "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": 19034.351299417205,
-      "required_pages_per_second_upper_estimate": 0.003964353406068845
+      "required_token_positions_per_second": 19236.90959388147,
+      "required_pages_per_second_upper_estimate": 0.004006540956984305
     },
     {
       "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": 28551.526949125808,
-      "required_pages_per_second_upper_estimate": 0.0017116204449427112
+      "required_token_positions_per_second": 28855.364390822207,
+      "required_pages_per_second_upper_estimate": 0.0017298350356394997
     },
     {
       "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": 28551.526949125808,
-      "required_pages_per_second_upper_estimate": 0.0022197556353997032
+      "required_token_positions_per_second": 28855.364390822207,
+      "required_pages_per_second_upper_estimate": 0.0022433776600519315
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28551.526949125808,
-      "required_pages_per_second_upper_estimate": 0.36622833146221584
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28855.364390822207,
+      "required_pages_per_second_upper_estimate": 0.3701256319290698
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.4184339780782125,
-    "producer_p95_seconds": 0.24920008999833954,
-    "gpu_update_seconds": 0.9683545139026803,
-    "producer_batches_per_second": 2.3898632816407726,
-    "gpu_batches_per_second": 1.032679649490951,
-    "producer_headroom": 2.3142348963871147,
+    "producer_batch_seconds": 0.1856973111486866,
+    "producer_p95_seconds": 0.24561469699983718,
+    "gpu_update_seconds": 0.9581580612024354,
+    "producer_batches_per_second": 5.385107591564999,
+    "gpu_batches_per_second": 1.0436691402930487,
+    "producer_headroom": 5.159784249300436,
     "sustainable": true,
-    "buffer_seconds": 60.03797986196618,
+    "buffer_seconds": 60.36395785575343,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 415533.0042156832,
-    "upload_bytes_per_second": 4981169.836414044,
+    "generation_bytes_per_second": 416958.6725018514,
+    "upload_bytes_per_second": 5001816.133671999,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.1637095914161,
+    "estimated_full_shards_per_hour": 89.46962481776863,
     "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": 1.5178640079993784,
-      "bytes": 12104268,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 0.9816815789999964,
       "bytes": 6359822,
@@ -562,6 +556,12 @@
       "bytes": 6624046,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.1070565920017543,
+      "bytes": 11831537,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -208,3 +208,4 @@
 {"step": 417900, "loss": 0.629663348197937, "lr": 0.0003, "grad_norm": 0.17760820649849354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61169583518, "tokens_per_second": 151079.328930287, "data_wait_s": 1.5313999028876424e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 418000, "loss": 0.5443013906478882, "lr": 0.0003, "grad_norm": 0.13174783431493042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61184254096, "tokens_per_second": 149585.14254806645, "data_wait_s": 1.0854000720428303e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 418100, "loss": 0.6076033115386963, "lr": 0.0003, "grad_norm": 0.1634113193470032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61198785933, "tokens_per_second": 150698.53566064817, "data_wait_s": 1.0841999028343707e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 418200, "loss": 0.5855352282524109, "lr": 0.0003, "grad_norm": 0.1395451328877831, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61213494863, "tokens_per_second": 152492.80641446833, "data_wait_s": 1.1574000382097438e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c7987174e4b79efaa17407f6597b377e3d12ce6a8ecfb8b943117d598f8cf9c3
-size 448077
+oid sha256:851b905adcc7c2b861d04cd43eceeb6ec96d7fee37a33ac78a9d797dcf097aad
+size 450195