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": 75740.10139994962,
-      "required_pages_per_second_upper_estimate": 0.005239875956096181
+      "required_token_positions_per_second": 75170.29934281763,
+      "required_pages_per_second_upper_estimate": 0.005200455727660864
     },
     {
       "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": 9467.512674993703,
-      "required_pages_per_second_upper_estimate": 0.003963849852873521
+      "required_token_positions_per_second": 9396.287417852203,
+      "required_pages_per_second_upper_estimate": 0.003934029325060877
     },
     {
       "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": 18935.025349987405,
-      "required_pages_per_second_upper_estimate": 0.0039436664302040505
+      "required_token_positions_per_second": 18792.574835704407,
+      "required_pages_per_second_upper_estimate": 0.003913997744751589
     },
     {
       "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": 28402.538024981106,
-      "required_pages_per_second_upper_estimate": 0.0017026887864331514
+      "required_token_positions_per_second": 28188.86225355661,
+      "required_pages_per_second_upper_estimate": 0.0016898792502002644
     },
     {
       "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": 28402.538024981106,
-      "required_pages_per_second_upper_estimate": 0.002208172401880472
+      "required_token_positions_per_second": 28188.86225355661,
+      "required_pages_per_second_upper_estimate": 0.002191560050512609
     },
     {
       "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": 28402.538024981106,
-      "required_pages_per_second_upper_estimate": 0.3643172615151307
+      "required_token_positions_per_second": 28188.86225355661,
+      "required_pages_per_second_upper_estimate": 0.36157645814646566
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11142892917197855,
-    "producer_p95_seconds": 0.2649414719999186,
-    "gpu_update_seconds": 0.973434133797569,
-    "producer_batches_per_second": 8.974330162112638,
-    "gpu_batches_per_second": 1.0272908718526153,
-    "producer_headroom": 8.735919307769514,
+    "producer_batch_seconds": 0.12993308771879697,
+    "producer_p95_seconds": 0.2752402260011877,
+    "gpu_update_seconds": 0.9808129094146618,
+    "producer_batches_per_second": 7.696269037831335,
+    "gpu_batches_per_second": 1.0195624368329215,
+    "producer_headroom": 7.5486000266333315,
     "sustainable": true,
-    "buffer_seconds": 60.35291629544928,
+    "buffer_seconds": 60.81040038370903,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287306.2932266875,
-    "upload_bytes_per_second": 5098154.501941984,
+    "generation_bytes_per_second": 328104.14742485655,
+    "upload_bytes_per_second": 5152756.524796958,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.64924237823934,
+    "estimated_full_shards_per_hour": 70.4035121637275,
     "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": 1.2874604959997669,
-      "bytes": 2530940,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.3693337599997903,
       "bytes": 10798536,
@@ -562,6 +556,12 @@
       "bytes": 1011694,
       "pages": 29,
       "shards": 1
+    },
+    {
+      "seconds": 1.350358371000766,
+      "bytes": 7369637,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -198,3 +198,4 @@
 {"step": 739800, "loss": 0.5362396240234375, "lr": 5e-05, "grad_norm": 0.23764093765216354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108286777059, "tokens_per_second": 148746.53036725926, "data_wait_s": 1.7947997548617423e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 739900, "loss": 0.5435524582862854, "lr": 5e-05, "grad_norm": 0.233901433441873, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108301331256, "tokens_per_second": 149060.28988581957, "data_wait_s": 2.0437997591216117e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 740000, "loss": 0.5588091611862183, "lr": 5e-05, "grad_norm": 0.2407953392202154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108315982560, "tokens_per_second": 150483.63865526207, "data_wait_s": 1.4635999832535163e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 740100, "loss": 0.5209000110626221, "lr": 5e-05, "grad_norm": 0.2159172759914041, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108330641559, "tokens_per_second": 139892.70958568153, "data_wait_s": 1.8845999875338748e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4836f0930117d1de75720fca0faab20346afd9b2276da3b56c82cc5aa4e8b0ec
-size 426897
+oid sha256:9e9e697a34faf01bbffa92ef6e6ed77db6f2d3e3f459cb070afbd653f7f68c73
+size 429015