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": 77244.7228259571,
-      "required_pages_per_second_upper_estimate": 0.005343969157550081
+      "required_token_positions_per_second": 77153.11974548269,
+      "required_pages_per_second_upper_estimate": 0.0053376318438944316
     },
     {
       "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": 9655.590353244637,
-      "required_pages_per_second_upper_estimate": 0.004042594049248654
+      "required_token_positions_per_second": 9644.139968185336,
+      "required_pages_per_second_upper_estimate": 0.004037800012135555
     },
     {
       "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": 19311.180706489275,
-      "required_pages_per_second_upper_estimate": 0.004022009671079556
+      "required_token_positions_per_second": 19288.279936370673,
+      "required_pages_per_second_upper_estimate": 0.004017240044597772
     },
     {
       "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": 28966.771059733914,
-      "required_pages_per_second_upper_estimate": 0.0017365136953326225
+      "required_token_positions_per_second": 28932.41990455601,
+      "required_pages_per_second_upper_estimate": 0.0017344543960308842
     },
     {
       "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": 28966.771059733914,
-      "required_pages_per_second_upper_estimate": 0.0022520390385336675
+      "required_token_positions_per_second": 28932.41990455601,
+      "required_pages_per_second_upper_estimate": 0.002249368387313349
     },
     {
       "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": 28966.771059733914,
-      "required_pages_per_second_upper_estimate": 0.3715546370587077
+      "required_token_positions_per_second": 28932.41990455601,
+      "required_pages_per_second_upper_estimate": 0.37111401732348237
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0826181027890911,
-    "producer_p95_seconds": 0.1725770940001894,
-    "gpu_update_seconds": 0.9544729698379427,
-    "producer_batches_per_second": 12.103884817505639,
-    "gpu_batches_per_second": 1.0476986060378297,
-    "producer_headroom": 11.552830888340992,
+    "producer_batch_seconds": 0.26597973821095167,
+    "producer_p95_seconds": 0.17864326700055244,
+    "gpu_update_seconds": 0.9556062054680138,
+    "producer_batches_per_second": 3.759684879480888,
+    "gpu_batches_per_second": 1.0464561597423325,
+    "producer_headroom": 3.5927782014361984,
     "sustainable": true,
-    "buffer_seconds": 60.13179709979039,
+    "buffer_seconds": 60.20319094448487,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 349872.5927126253,
-    "upload_bytes_per_second": 5730174.961226118,
+    "generation_bytes_per_second": 367454.253725987,
+    "upload_bytes_per_second": 5673195.403291617,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.07451377901143,
+    "estimated_full_shards_per_hour": 78.84712895235737,
     "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": [
@@ -466,6 +466,12 @@
       "bytes": 9869427,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.9875589590001255,
+      "bytes": 2599753,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -56,3 +56,4 @@
 {"step": 562700, "loss": 0.5789791345596313, "lr": 0.0003, "grad_norm": 0.1317799438170744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82362214380, "tokens_per_second": 142841.35138604647, "data_wait_s": 1.5511999663431197e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 562800, "loss": 0.5476948618888855, "lr": 0.0003, "grad_norm": 0.15059537639775994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82376740089, "tokens_per_second": 151489.53378136404, "data_wait_s": 1.3345000297704246e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 562900, "loss": 0.5054783821105957, "lr": 0.0003, "grad_norm": 0.1824266095234045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82391282928, "tokens_per_second": 151577.83447237167, "data_wait_s": 1.6808000509627163e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 563000, "loss": 0.6174187064170837, "lr": 0.0003, "grad_norm": 0.16749621556641014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82405831092, "tokens_per_second": 151664.65638982385, "data_wait_s": 1.306000012846198e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d274482a991117d489653df0b501eee8ad43cd12f81cfe873492c91834f8d2e0
-size 123792
+oid sha256:629b422d888d97848fb64a8b8bc904f7fcb4982b6a714eda89f1282cec89d923
+size 125910