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": 76520.20411134923,
-      "required_pages_per_second_upper_estimate": 0.005293845271758479
+      "required_token_positions_per_second": 76928.83511951414,
+      "required_pages_per_second_upper_estimate": 0.005322115313057896
     },
     {
       "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": 9565.025513918654,
-      "required_pages_per_second_upper_estimate": 0.004004676442231766
+      "required_token_positions_per_second": 9616.104389939268,
+      "required_pages_per_second_upper_estimate": 0.004026062100973373
     },
     {
       "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": 19130.051027837308,
-      "required_pages_per_second_upper_estimate": 0.0039842851357271974
+      "required_token_positions_per_second": 19232.208779878536,
+      "required_pages_per_second_upper_estimate": 0.004005561901396295
     },
     {
       "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": 28695.076541755963,
-      "required_pages_per_second_upper_estimate": 0.0017202260238333527
+      "required_token_positions_per_second": 28848.313169817804,
+      "required_pages_per_second_upper_estimate": 0.0017294123257068767
     },
     {
       "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": 28695.076541755963,
-      "required_pages_per_second_upper_estimate": 0.002230915984818769
+      "required_token_positions_per_second": 28848.313169817804,
+      "required_pages_per_second_upper_estimate": 0.0022428294586338826
     },
     {
       "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": 28695.076541755963,
-      "required_pages_per_second_upper_estimate": 0.3680696315049315
+      "required_token_positions_per_second": 28848.313169817804,
+      "required_pages_per_second_upper_estimate": 0.37003518643703653
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4656333128515513,
-    "producer_p95_seconds": 0.20321868599967274,
-    "gpu_update_seconds": 0.9635102370181067,
-    "producer_batches_per_second": 2.1476126651591407,
-    "gpu_batches_per_second": 1.037871692048465,
-    "producer_headroom": 2.069246788030571,
+    "producer_batch_seconds": 0.10127828546096396,
+    "producer_p95_seconds": 0.18950245100131724,
+    "gpu_update_seconds": 0.9583922580584844,
+    "producer_batches_per_second": 9.873784843893645,
+    "gpu_batches_per_second": 1.0434141048111185,
+    "producer_headroom": 9.46295895212287,
     "sustainable": true,
-    "buffer_seconds": 60.70114493214072,
+    "buffer_seconds": 60.37871225768452,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 304167.37147786235,
-    "upload_bytes_per_second": 4033461.702671087,
+    "generation_bytes_per_second": 304379.6520302227,
+    "upload_bytes_per_second": 4061027.5664061466,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.26723726512816,
+    "estimated_full_shards_per_hour": 65.31278773002634,
     "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": [
@@ -394,6 +394,12 @@
       "bytes": 6693273,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.380750825999712,
+      "bytes": 7212989,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -83,3 +83,4 @@
 {"step": 808300, "loss": 0.5780943036079407, "lr": 3e-05, "grad_norm": 0.3202913211437767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118311574940, "tokens_per_second": 151289.2426151579, "data_wait_s": 9.577999662724324e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 808400, "loss": 0.5267443656921387, "lr": 3e-05, "grad_norm": 0.2612070844248897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118326353095, "tokens_per_second": 132708.07914494834, "data_wait_s": 1.7387999832862988e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 808500, "loss": 0.5446516871452332, "lr": 3e-05, "grad_norm": 0.26892138771215574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118341033127, "tokens_per_second": 152401.82121693698, "data_wait_s": 1.0585001291474327e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 808600, "loss": 0.5144476294517517, "lr": 3e-05, "grad_norm": 0.2696697976138345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118355747682, "tokens_per_second": 145440.98551701644, "data_wait_s": 9.385001249029301e-06, "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 sha256:57a8973f04735d4a1326c273ab080cb288f6052b96a92a870e406457ef24228e
-size 180978
+oid sha256:170e4dbd1091f4c196971b3eeba5902131d69b326f52adc4135b99bc4aebbed0
+size 183096