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": 77454.60382192247,
-      "required_pages_per_second_upper_estimate": 0.00535848921184197
+      "required_token_positions_per_second": 76476.76156362565,
+      "required_pages_per_second_upper_estimate": 0.005290839815506363
     },
     {
       "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": 9681.825477740309,
-      "required_pages_per_second_upper_estimate": 0.004053578148023294
+      "required_token_positions_per_second": 9559.595195453207,
+      "required_pages_per_second_upper_estimate": 0.0040024028812882285
     },
     {
       "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": 19363.650955480618,
-      "required_pages_per_second_upper_estimate": 0.004032937840210935
+      "required_token_positions_per_second": 19119.190390906413,
+      "required_pages_per_second_upper_estimate": 0.0039820231514687485
     },
     {
       "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": 29045.476433220927,
-      "required_pages_per_second_upper_estimate": 0.001741231962297022
+      "required_token_positions_per_second": 28678.785586359616,
+      "required_pages_per_second_upper_estimate": 0.0017192494059321902
     },
     {
       "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": 29045.476433220927,
-      "required_pages_per_second_upper_estimate": 0.0022581580351339315
+      "required_token_positions_per_second": 28678.785586359616,
+      "required_pages_per_second_upper_estimate": 0.002229649434693043
     },
     {
       "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": 29045.476433220927,
-      "required_pages_per_second_upper_estimate": 0.3725641850825532
+      "required_token_positions_per_second": 28678.785586359616,
+      "required_pages_per_second_upper_estimate": 0.3678606686209722
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09667991869531534,
-    "producer_p95_seconds": 0.2657777909989818,
-    "gpu_update_seconds": 0.9518866066310224,
-    "producer_batches_per_second": 10.343409608684905,
-    "gpu_batches_per_second": 1.0505452992339745,
-    "producer_headroom": 9.845753073405785,
+    "producer_batch_seconds": 0.09557864613276479,
+    "producer_p95_seconds": 0.26468162899982417,
+    "gpu_update_seconds": 0.9640575580421409,
+    "producer_batches_per_second": 10.462588040961961,
+    "gpu_batches_per_second": 1.037282464784419,
+    "producer_headroom": 10.086537077570695,
     "sustainable": true,
-    "buffer_seconds": 60.920742824385435,
+    "buffer_seconds": 60.735626156654874,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331935.59637974895,
-    "upload_bytes_per_second": 4589739.18727165,
+    "generation_bytes_per_second": 296909.81324083416,
+    "upload_bytes_per_second": 4641064.449108352,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.22565191788055,
+    "estimated_full_shards_per_hour": 63.709934214770975,
     "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": 1.2175075589984772,
-      "bytes": 2434158,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.370172322000144,
       "bytes": 11210739,
@@ -562,6 +556,12 @@
       "bytes": 2384330,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0045538510021288,
+      "bytes": 5798532,
+      "pages": 25,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -287,3 +287,4 @@
 {"step": 713700, "loss": 0.6320292949676514, "lr": 8e-05, "grad_norm": 0.39268109388291367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104465893600, "tokens_per_second": 153156.06101313586, "data_wait_s": 1.5458997950190678e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 713800, "loss": 0.5295023322105408, "lr": 8e-05, "grad_norm": 0.21661962613630112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104480523940, "tokens_per_second": 152592.07676565228, "data_wait_s": 1.0264000593451783e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 713900, "loss": 0.5419623255729675, "lr": 8e-05, "grad_norm": 0.21912230974499358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104495133919, "tokens_per_second": 143513.54363003792, "data_wait_s": 1.5841000276850536e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 714000, "loss": 0.5525996088981628, "lr": 8e-05, "grad_norm": 0.22174559780674935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104509828971, "tokens_per_second": 152933.54303830554, "data_wait_s": 1.1226999049540609e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:12622aa3226d361bf65a9759931deb110d29e9967474220f31c43d6df85024ad
-size 616182
+oid sha256:945ea83a121b53fab1778954b269282333746638b0785e72882a2646947b9065
+size 618300