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": 77650.72122925837,
-      "required_pages_per_second_upper_estimate": 0.005372057069136548
+      "required_token_positions_per_second": 77047.87455007818,
+      "required_pages_per_second_upper_estimate": 0.005330350737073849
     },
     {
       "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": 9706.340153657296,
-      "required_pages_per_second_upper_estimate": 0.00406384193090509
+      "required_token_positions_per_second": 9630.984318759773,
+      "required_pages_per_second_upper_estimate": 0.004032292016442271
     },
     {
       "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": 19412.680307314593,
-      "required_pages_per_second_upper_estimate": 0.004043149361206994
+      "required_token_positions_per_second": 19261.968637519545,
+      "required_pages_per_second_upper_estimate": 0.004011760094922694
     },
     {
       "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": 29119.02046097189,
-      "required_pages_per_second_upper_estimate": 0.001745640814465464
+      "required_token_positions_per_second": 28892.952956279318,
+      "required_pages_per_second_upper_estimate": 0.0017320884127442371
     },
     {
       "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": 29119.02046097189,
-      "required_pages_per_second_upper_estimate": 0.0022638757598055777
+      "required_token_positions_per_second": 28892.952956279318,
+      "required_pages_per_second_upper_estimate": 0.0022463000056815955
     },
     {
       "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": 29119.02046097189,
-      "required_pages_per_second_upper_estimate": 0.3735075289051178
+      "required_token_positions_per_second": 28892.952956279318,
+      "required_pages_per_second_upper_estimate": 0.3706077776872964
     }
   ],
   "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": 0.09045546167180873,
-    "producer_p95_seconds": 0.3170166869967943,
-    "gpu_update_seconds": 0.9494824881577493,
-    "producer_batches_per_second": 11.055164403761582,
-    "gpu_batches_per_second": 1.0532053118117726,
-    "producer_headroom": 10.496685005076527,
+    "producer_batch_seconds": 0.1454622290393104,
+    "producer_p95_seconds": 0.36866725900472375,
+    "gpu_update_seconds": 0.9569115362433471,
+    "producer_batches_per_second": 6.874636849747128,
+    "gpu_batches_per_second": 1.0450286804209823,
+    "producer_headroom": 6.578419309006649,
     "sustainable": true,
-    "buffer_seconds": 60.76687924209595,
+    "buffer_seconds": 60.28542678333087,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362460.5082247594,
-    "upload_bytes_per_second": 5041760.896956847,
+    "generation_bytes_per_second": 387454.9196368187,
+    "upload_bytes_per_second": 4989202.019505149,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.77558741623962,
+    "estimated_full_shards_per_hour": 83.13880626514836,
     "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": 1.4021912890020758,
-      "bytes": 11567150,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 1.280043183000089,
       "bytes": 12206708,
@@ -562,6 +556,12 @@
       "bytes": 5675347,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.561844530995586,
+      "bytes": 7766796,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -406,3 +406,4 @@
 {"step": 685600, "loss": 0.5104237198829651, "lr": 0.0001, "grad_norm": 0.19530410664371448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100354050683, "tokens_per_second": 151499.80442519503, "data_wait_s": 1.3083001249469817e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 685700, "loss": 0.5354567170143127, "lr": 0.0001, "grad_norm": 0.20256810840400116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100368548106, "tokens_per_second": 151211.0827327131, "data_wait_s": 1.2427000910975039e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 685800, "loss": 0.5752155184745789, "lr": 0.0001, "grad_norm": 0.22514589126917264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100383214867, "tokens_per_second": 152487.6991032466, "data_wait_s": 1.1365002137608826e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 685900, "loss": 0.513319730758667, "lr": 0.0001, "grad_norm": 0.19845639430238482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100397882630, "tokens_per_second": 146969.60658545097, "data_wait_s": 2.0419996872078627e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:43627c73d7aac8b517a28578c5fb58c234f82a6e5558e609345c45728c9fbaff
-size 870573
+oid sha256:fee26e3229df71b0b06840d944a42ad0c08fb76fd75e063c7369aca445b9dbe9
+size 872691