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": 77267.78512016054,
-      "required_pages_per_second_upper_estimate": 0.00534556466057497
+      "required_token_positions_per_second": 77100.1333789849,
+      "required_pages_per_second_upper_estimate": 0.005333966124114805
     },
     {
       "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": 9658.473140020067,
-      "required_pages_per_second_upper_estimate": 0.004043801011872093
+      "required_token_positions_per_second": 9637.516672373113,
+      "required_pages_per_second_upper_estimate": 0.004035026976489121
     },
     {
       "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": 19316.946280040134,
-      "required_pages_per_second_upper_estimate": 0.004023210488001801
+      "required_token_positions_per_second": 19275.033344746225,
+      "required_pages_per_second_upper_estimate": 0.004014481128898513
     },
     {
       "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": 28975.419420060203,
-      "required_pages_per_second_upper_estimate": 0.001737032151328902
+      "required_token_positions_per_second": 28912.55001711934,
+      "required_pages_per_second_upper_estimate": 0.0017332632266186172
     },
     {
       "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": 28975.419420060203,
-      "required_pages_per_second_upper_estimate": 0.00225271141050892
+      "required_token_positions_per_second": 28912.55001711934,
+      "required_pages_per_second_upper_estimate": 0.0022478235909635462
     },
     {
       "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": 28975.419420060203,
-      "required_pages_per_second_upper_estimate": 0.37166556893908753
+      "required_token_positions_per_second": 28912.55001711934,
+      "required_pages_per_second_upper_estimate": 0.37085914774206774
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12158256860925576,
-    "producer_p95_seconds": 0.3381743370046024,
-    "gpu_update_seconds": 0.954188086087161,
-    "producer_batches_per_second": 8.224863246752237,
-    "gpu_batches_per_second": 1.0480114084223162,
-    "producer_headroom": 7.84806651974715,
+    "producer_batch_seconds": 0.11323859389830204,
+    "producer_p95_seconds": 0.34007354699861025,
+    "gpu_update_seconds": 0.9562629371546062,
+    "producer_batches_per_second": 8.830911490282947,
+    "gpu_batches_per_second": 1.0457374861515965,
+    "producer_headroom": 8.444673359450332,
     "sustainable": true,
-    "buffer_seconds": 60.11384942349114,
+    "buffer_seconds": 60.24456504074019,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396155.61459481553,
-    "upload_bytes_per_second": 4831960.809389454,
+    "generation_bytes_per_second": 419450.9059471112,
+    "upload_bytes_per_second": 4762350.11513417,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.00577286132192,
+    "estimated_full_shards_per_hour": 90.00440009889604,
     "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": 1.4905629579989181,
-      "bytes": 13725537,
-      "pages": 44,
-      "shards": 4
-    },
     {
       "seconds": 1.3237662440005806,
       "bytes": 894969,
@@ -562,6 +556,12 @@
       "bytes": 11067080,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.1875661560043227,
+      "bytes": 6040644,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -352,3 +352,4 @@
 {"step": 960200, "loss": 0.5323472619056702, "lr": 3e-05, "grad_norm": 0.3651668859079228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140547305755, "tokens_per_second": 152614.19836806075, "data_wait_s": 1.561100361868739e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 960300, "loss": 0.4896109700202942, "lr": 3e-05, "grad_norm": 0.3447349619880113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140561809848, "tokens_per_second": 150716.92197168712, "data_wait_s": 1.5595003787893802e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 960400, "loss": 0.5473693013191223, "lr": 3e-05, "grad_norm": 0.3466411055769713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140576307698, "tokens_per_second": 150334.93396943517, "data_wait_s": 1.3853998098056763e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 960500, "loss": 0.5665472149848938, "lr": 3e-05, "grad_norm": 0.3649950479541535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140590879064, "tokens_per_second": 151317.5549284369, "data_wait_s": 1.82330040843226e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c5d88341c4f82552637f836297fe42724b9aeb4eeff50f992183d89b1c7c5686
-size 755418
+oid sha256:7451543400c1983d6cc9496482554c0b9e4482d1a243cde1972625b6df8fa33e
+size 757536