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": 77479.03589029987,
-      "required_pages_per_second_upper_estimate": 0.005360179479022528
+      "required_token_positions_per_second": 76997.21405326622,
+      "required_pages_per_second_upper_estimate": 0.00532684592635585
     },
     {
       "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": 9684.879486287484,
-      "required_pages_per_second_upper_estimate": 0.004054856797627046
+      "required_token_positions_per_second": 9624.651756658277,
+      "required_pages_per_second_upper_estimate": 0.004029640704929309
     },
     {
       "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": 19369.758972574968,
-      "required_pages_per_second_upper_estimate": 0.004034209979092445
+      "required_token_positions_per_second": 19249.303513316554,
+      "required_pages_per_second_upper_estimate": 0.004009122283552997
     },
     {
       "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": 29054.63845886245,
-      "required_pages_per_second_upper_estimate": 0.0017417812117446283
+      "required_token_positions_per_second": 28873.955269974827,
+      "required_pages_per_second_upper_estimate": 0.0017309495304580703
     },
     {
       "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": 29054.63845886245,
-      "required_pages_per_second_upper_estimate": 0.002258870342328068
+      "required_token_positions_per_second": 28873.955269974827,
+      "required_pages_per_second_upper_estimate": 0.0022448230191334123
     },
     {
       "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": 29054.63845886245,
-      "required_pages_per_second_upper_estimate": 0.3726817057100659
+      "required_token_positions_per_second": 28873.955269974827,
+      "required_pages_per_second_upper_estimate": 0.37036409576550877
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1601542736796091,
-    "producer_p95_seconds": 0.21718408000015188,
-    "gpu_update_seconds": 0.9515864408068934,
-    "producer_batches_per_second": 6.243979489429762,
-    "gpu_batches_per_second": 1.0508766803697356,
-    "producer_headroom": 5.94168621881771,
+    "producer_batch_seconds": 0.3665661124764341,
+    "producer_p95_seconds": 0.21266951899997366,
+    "gpu_update_seconds": 0.9575411384234683,
+    "producer_batches_per_second": 2.7280208561676256,
+    "gpu_batches_per_second": 1.044341553456844,
+    "producer_headroom": 2.612192196257713,
     "sustainable": true,
-    "buffer_seconds": 60.901532211641175,
+    "buffer_seconds": 60.32509172067851,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382319.1662970033,
-    "upload_bytes_per_second": 5358628.355784241,
+    "generation_bytes_per_second": 346371.1704702944,
+    "upload_bytes_per_second": 5436254.051370593,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.03679315264296,
+    "estimated_full_shards_per_hour": 74.32319007474541,
     "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": 1.1117148500006806,
-      "bytes": 6299752,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 3.5441821820004407,
       "bytes": 11982351,
@@ -562,6 +556,12 @@
       "bytes": 1252571,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.2397414250008296,
+      "bytes": 13176621,
+      "pages": 29,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -147,3 +147,4 @@
 {"step": 571800, "loss": 0.5821009278297424, "lr": 0.0003, "grad_norm": 0.16956327297832322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83694621871, "tokens_per_second": 150847.4240168994, "data_wait_s": 1.1878000805154443e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 571900, "loss": 0.5540379881858826, "lr": 0.0003, "grad_norm": 0.17463754712443672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83709282993, "tokens_per_second": 153536.64720869836, "data_wait_s": 5.871999746887013e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 572000, "loss": 0.5810781717300415, "lr": 0.0003, "grad_norm": 0.16302066896464115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83723947374, "tokens_per_second": 153297.34922635122, "data_wait_s": 8.31400029710494e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 572100, "loss": 0.5820192694664001, "lr": 0.0003, "grad_norm": 0.19463367989754726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83738710562, "tokens_per_second": 153748.82771875543, "data_wait_s": 1.507500019215513e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1797c02250a3132cfd542ff52841996ddff74697035c7d3708d18eb202170423
-size 318096
+oid sha256:9b1858390f633494af33dd3a914b2457d51134cd095063575047735136ed72e9
+size 320214