Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74057.62506718993,
-      "required_pages_per_second_upper_estimate": 0.006419127243489303
+      "required_token_positions_per_second": 74777.18962873796,
+      "required_pages_per_second_upper_estimate": 0.006481497275964579
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9257.203133398742,
-      "required_pages_per_second_upper_estimate": 0.003692836737433677
+      "required_token_positions_per_second": 9347.148703592246,
+      "required_pages_per_second_upper_estimate": 0.0037287173701899814
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18514.406266797483,
-      "required_pages_per_second_upper_estimate": 0.004282982047402202
+      "required_token_positions_per_second": 18694.29740718449,
+      "required_pages_per_second_upper_estimate": 0.004324596696754798
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27771.609400196223,
-      "required_pages_per_second_upper_estimate": 0.001691145306098209
+      "required_token_positions_per_second": 28041.446110776735,
+      "required_pages_per_second_upper_estimate": 0.0017075769460487547
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27771.609400196223,
-      "required_pages_per_second_upper_estimate": 0.002137935368362255
+      "required_token_positions_per_second": 28041.446110776735,
+      "required_pages_per_second_upper_estimate": 0.002158708145298565
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27771.609400196223,
-      "required_pages_per_second_upper_estimate": 0.3448304431527897
+      "required_token_positions_per_second": 28041.446110776735,
+      "required_pages_per_second_upper_estimate": 0.3481809120128231
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2713564281873744,
-    "producer_p95_seconds": 0.3436247520003235,
-    "gpu_update_seconds": 0.9955490732130436,
-    "producer_batches_per_second": 3.6851900162449436,
-    "gpu_batches_per_second": 1.0044708261066342,
-    "producer_headroom": 3.668787505286615,
+    "producer_batch_seconds": 0.07501082764048306,
+    "producer_p95_seconds": 0.304767540001194,
+    "gpu_update_seconds": 0.9859691219481891,
+    "producer_batches_per_second": 13.331408697326568,
+    "gpu_batches_per_second": 1.0142305450946447,
+    "producer_headroom": 13.144357327635529,
     "sustainable": true,
-    "buffer_seconds": 60.72849346599566,
+    "buffer_seconds": 60.14411643883953,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421983.2956566713,
-    "upload_bytes_per_second": 5489352.199222883,
+    "generation_bytes_per_second": 405105.5905942224,
+    "upload_bytes_per_second": 5540009.560119354,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.5477919795523,
+    "estimated_full_shards_per_hour": 86.92622936601643,
     "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": 1.121872658999564,
-      "bytes": 7611150,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.0686068680006429,
       "bytes": 1192521,
@@ -562,6 +556,12 @@
       "bytes": 949451,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.1886153279992868,
+      "bytes": 11793828,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -224,3 +224,4 @@
 {"step": 177400, "loss": 0.5940247401595116, "lr": 0.0003, "grad_norm": 0.13568014934017145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25963575225, "tokens_per_second": 149906.75099345375, "data_wait_s": 1.5060002624522895e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 177500, "loss": 0.5926298797130585, "lr": 0.0003, "grad_norm": 0.14159742652442084, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25978061501, "tokens_per_second": 146471.00631634385, "data_wait_s": 1.5407000319100916e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 177600, "loss": 0.5881208218634129, "lr": 0.0003, "grad_norm": 0.1575129915447211, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25992501304, "tokens_per_second": 147525.08178910555, "data_wait_s": 1.1316002201056108e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 177700, "loss": 0.5706491265445948, "lr": 0.0003, "grad_norm": 0.15813942364198907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26007069522, "tokens_per_second": 147803.53629690854, "data_wait_s": 1.3288001355249435e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a15429eef75ad0a6b2548fe28558808785f22fdd69ca59c1df8aaac729360f81
-size 481965
+oid sha256:0833f3476863e94fbfd5644ed7d0fba50e38a6a714e58e76b7b6bc4ff1ffd513
+size 484083