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": 74701.31975085931,
-      "required_pages_per_second_upper_estimate": 0.005168010631306613
+      "required_token_positions_per_second": 75641.7297664733,
+      "required_pages_per_second_upper_estimate": 0.005233070378238653
     },
     {
       "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": 9337.664968857414,
-      "required_pages_per_second_upper_estimate": 0.0039094853298427
+      "required_token_positions_per_second": 9455.216220809163,
+      "required_pages_per_second_upper_estimate": 0.0039587015842856175
     },
     {
       "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": 18675.32993771483,
-      "required_pages_per_second_upper_estimate": 0.0038895787244562453
+      "required_token_positions_per_second": 18910.432441618326,
+      "required_pages_per_second_upper_estimate": 0.003938544375949379
     },
     {
       "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": 28012.994906572243,
-      "required_pages_per_second_upper_estimate": 0.0016793362712824418
+      "required_token_positions_per_second": 28365.648662427488,
+      "required_pages_per_second_upper_estimate": 0.001700477325474857
     },
     {
       "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": 28012.994906572243,
-      "required_pages_per_second_upper_estimate": 0.0021778871378432814
+      "required_token_positions_per_second": 28365.648662427488,
+      "required_pages_per_second_upper_estimate": 0.002205304416201086
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28012.994906572243,
-      "required_pages_per_second_upper_estimate": 0.3593206206509953
+      "required_token_positions_per_second": 28365.648662427488,
+      "required_pages_per_second_upper_estimate": 0.36384408438100446
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1499849771253139,
-    "producer_p95_seconds": 0.37857038999936776,
-    "gpu_update_seconds": 0.9869705146561601,
-    "producer_batches_per_second": 6.667334416862899,
-    "gpu_batches_per_second": 1.0132014940166465,
-    "producer_headroom": 6.580462480795905,
+    "producer_batch_seconds": 0.09976324890646993,
+    "producer_p95_seconds": 0.35383229500439484,
+    "gpu_update_seconds": 0.974700079276591,
+    "producer_batches_per_second": 10.023731293449758,
+    "gpu_batches_per_second": 1.0259566211815498,
+    "producer_headroom": 9.770131686372725,
     "sustainable": true,
-    "buffer_seconds": 60.20520139402577,
+    "buffer_seconds": 60.43140491514864,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298626.1227862486,
-    "upload_bytes_per_second": 4696556.373573309,
+    "generation_bytes_per_second": 298861.00976301555,
+    "upload_bytes_per_second": 4654394.899486298,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.07821428957553,
+    "estimated_full_shards_per_hour": 64.12861556690073,
     "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": 1.3368553659965983,
-      "bytes": 12766871,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.6370693080025376,
       "bytes": 5715454,
@@ -562,6 +556,12 @@
       "bytes": 5876405,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0105084750030073,
+      "bytes": 7339502,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -429,3 +429,4 @@
 {"step": 687900, "loss": 0.5393136739730835, "lr": 0.0001, "grad_norm": 0.23868798620330395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100690033392, "tokens_per_second": 150711.7483638031, "data_wait_s": 1.4796998584643006e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688000, "loss": 0.4782137870788574, "lr": 0.0001, "grad_norm": 0.20103014709600134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100704686333, "tokens_per_second": 151602.26420412486, "data_wait_s": 1.779800368240103e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688100, "loss": 0.5473445057868958, "lr": 0.0001, "grad_norm": 0.20208139775954997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100719252292, "tokens_per_second": 151831.41347636824, "data_wait_s": 9.546994988340884e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 688200, "loss": 0.5151166915893555, "lr": 0.0001, "grad_norm": 0.2120875927685729, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100733890275, "tokens_per_second": 153225.4642736849, "data_wait_s": 1.7035999917425215e-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:b578d4a82d34b84a19b5c223c67f79af1b2c94f05ea8541f30140e3c7bcad565
-size 919287
+oid sha256:5037f035d3e12b4e87ddcd141b8cd548fef149b7f6c442101c639c1c4dba7aa4
+size 921405