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": 76041.5479544816,
-      "required_pages_per_second_upper_estimate": 0.005260730728191076
+      "required_token_positions_per_second": 76075.92805648268,
+      "required_pages_per_second_upper_estimate": 0.005263109223420338
     },
     {
       "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": 9505.1934943102,
-      "required_pages_per_second_upper_estimate": 0.0039796260250563546
+      "required_token_positions_per_second": 9509.491007060335,
+      "required_pages_per_second_upper_estimate": 0.003981425303902565
     },
     {
       "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": 19010.3869886204,
-      "required_pages_per_second_upper_estimate": 0.003959362272111245
+      "required_token_positions_per_second": 19018.98201412067,
+      "required_pages_per_second_upper_estimate": 0.003961152389256886
     },
     {
       "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": 28515.5804829306,
-      "required_pages_per_second_upper_estimate": 0.0017094655091813844
+      "required_token_positions_per_second": 28528.473021181006,
+      "required_pages_per_second_upper_estimate": 0.00171023839716373
     },
     {
       "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": 28515.5804829306,
-      "required_pages_per_second_upper_estimate": 0.0022169609557648206
+      "required_token_positions_per_second": 28528.473021181006,
+      "required_pages_per_second_upper_estimate": 0.002217963293905507
     },
     {
       "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": 28515.5804829306,
-      "required_pages_per_second_upper_estimate": 0.36576724879017203
+      "required_token_positions_per_second": 28528.473021181006,
+      "required_pages_per_second_upper_estimate": 0.3659326204279192
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12342861519513804,
-    "producer_p95_seconds": 0.2880671969978721,
-    "gpu_update_seconds": 0.9695752122791281,
-    "producer_batches_per_second": 8.101848979015289,
-    "gpu_batches_per_second": 1.0313795024208117,
-    "producer_headroom": 7.8553519436821855,
+    "producer_batch_seconds": 0.07644168762513459,
+    "producer_p95_seconds": 0.2796867309989466,
+    "gpu_update_seconds": 0.9691370435239455,
+    "producer_batches_per_second": 13.081867120777599,
+    "gpu_batches_per_second": 1.0318458123980399,
+    "producer_headroom": 12.67812202520351,
     "sustainable": true,
-    "buffer_seconds": 60.113663161305944,
+    "buffer_seconds": 60.086496698484616,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 467089.44112897076,
+    "generation_bytes_per_second": 495437.6219621759,
     "upload_bytes_per_second": 3282361.4383804593,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.22652078058093,
+    "estimated_full_shards_per_hour": 106.30938047551115,
     "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": [

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -258,3 +258,4 @@
 {"step": 825800, "loss": 0.5232481360435486, "lr": 3e-05, "grad_norm": 0.29386732985607794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120873850133, "tokens_per_second": 151178.61603742206, "data_wait_s": 2.2290998458629474e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 825900, "loss": 0.5090745091438293, "lr": 3e-05, "grad_norm": 0.2771079545371284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120888472424, "tokens_per_second": 150278.06130885956, "data_wait_s": 1.1221000022487715e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 826000, "loss": 0.5335186719894409, "lr": 3e-05, "grad_norm": 0.27717066949513663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120903085084, "tokens_per_second": 151215.27428752833, "data_wait_s": 9.088002116186544e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 826100, "loss": 0.49807727336883545, "lr": 3e-05, "grad_norm": 0.2930168926589278, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120917748156, "tokens_per_second": 151171.80290622264, "data_wait_s": 1.2401997082633898e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:87be1fdd90003f6993e09da6b9aa51c49a378c556a4b7db472eaad6c6779ddc1
-size 554760
+oid sha256:55a76eb3504850020ead90986f9bc3fe6082db1d7e1d00bbb5fbb37fb76958d3
+size 556878