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": 77124.77075958335,
-      "required_pages_per_second_upper_estimate": 0.005335670595271205
+      "required_token_positions_per_second": 76309.52882309728,
+      "required_pages_per_second_upper_estimate": 0.005279270266483191
     },
     {
       "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": 9640.596344947919,
-      "required_pages_per_second_upper_estimate": 0.004036316371085316
+      "required_token_positions_per_second": 9538.69110288716,
+      "required_pages_per_second_upper_estimate": 0.003993650774257916
     },
     {
       "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": 19281.192689895837,
-      "required_pages_per_second_upper_estimate": 0.004015763958060322
+      "required_token_positions_per_second": 19077.38220577432,
+      "required_pages_per_second_upper_estimate": 0.003973315609061719
     },
     {
       "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": 28921.789034843758,
-      "required_pages_per_second_upper_estimate": 0.0017338170916240327
+      "required_token_positions_per_second": 28616.073308661482,
+      "required_pages_per_second_upper_estimate": 0.0017154899006402906
     },
     {
       "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": 28921.789034843758,
-      "required_pages_per_second_upper_estimate": 0.0022485418839534757
+      "required_token_positions_per_second": 28616.073308661482,
+      "required_pages_per_second_upper_estimate": 0.002224773830944171
     },
     {
       "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": 28921.789034843758,
-      "required_pages_per_second_upper_estimate": 0.3709776559413522
+      "required_token_positions_per_second": 28616.073308661482,
+      "required_pages_per_second_upper_estimate": 0.3670562628578582
     }
   ],
   "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": 0.12078781743787204,
-    "producer_p95_seconds": 0.3697242540001753,
-    "gpu_update_seconds": 0.9559574605392097,
-    "producer_batches_per_second": 8.278980622481702,
-    "gpu_batches_per_second": 1.0460716520125781,
-    "producer_headroom": 7.914353291720934,
+    "producer_batch_seconds": 0.32477090442944245,
+    "producer_p95_seconds": 0.3489179480020539,
+    "gpu_update_seconds": 0.9661702953364861,
+    "producer_batches_per_second": 3.0790935590637347,
+    "gpu_batches_per_second": 1.0350142255736936,
+    "producer_headroom": 2.9749287333292807,
     "sustainable": true,
-    "buffer_seconds": 60.22532001397021,
+    "buffer_seconds": 60.86872860619862,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 464511.2304867355,
+    "generation_bytes_per_second": 434152.3910179972,
     "upload_bytes_per_second": 5495858.945531176,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.67329679442929,
+    "estimated_full_shards_per_hour": 93.15899656205117,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -424,3 +424,4 @@
 {"step": 599500, "loss": 0.5904302597045898, "lr": 0.0003, "grad_norm": 0.1590978193353224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87749038407, "tokens_per_second": 150917.65901102405, "data_wait_s": 1.0568997822701931e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 599600, "loss": 0.6239014267921448, "lr": 0.0003, "grad_norm": 0.1821307780857641, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87763667203, "tokens_per_second": 151769.4706270516, "data_wait_s": 1.1385003745090216e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 599700, "loss": 0.5784598588943481, "lr": 0.0003, "grad_norm": 0.15435013118863627, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87778325021, "tokens_per_second": 151965.5059810186, "data_wait_s": 1.3475000741891563e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 599800, "loss": 0.5953309535980225, "lr": 0.0003, "grad_norm": 0.21297484014685433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87792955888, "tokens_per_second": 151186.3068775913, "data_wait_s": 1.3890996342524886e-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 sha256:3cf5558d3323bddca81f07e8cd5e705589c5b304cac500c04e8b37e8e6601ab3
-size 908697
+oid sha256:7cbc72ce2fe019d915385ea7a8cbeb98b5ab8242d95dcf326483e1f5cf7a7e0f
+size 910815