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": 73410.80307904552,
-      "required_pages_per_second_upper_estimate": 0.006363062352912287
+      "required_token_positions_per_second": 74284.64655304699,
+      "required_pages_per_second_upper_estimate": 0.006438804890502697
     },
     {
       "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": 9176.35038488069,
-      "required_pages_per_second_upper_estimate": 0.0036605833671935098
+      "required_token_positions_per_second": 9285.580819130873,
+      "required_pages_per_second_upper_estimate": 0.003704157020556436
     },
     {
       "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": 18352.70076976138,
-      "required_pages_per_second_upper_estimate": 0.004245574326582448
+      "required_token_positions_per_second": 18571.161638261747,
+      "required_pages_per_second_upper_estimate": 0.004296111403730037
     },
     {
       "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": 27529.051154642068,
-      "required_pages_per_second_upper_estimate": 0.0016763747815487233
+      "required_token_positions_per_second": 27856.742457392622,
+      "required_pages_per_second_upper_estimate": 0.0016963294626228382
     },
     {
       "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": 27529.051154642068,
-      "required_pages_per_second_upper_estimate": 0.002119262563175298
+      "required_token_positions_per_second": 27856.742457392622,
+      "required_pages_per_second_upper_estimate": 0.0021444891467686193
     },
     {
       "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": 27529.051154642068,
-      "required_pages_per_second_upper_estimate": 0.341818681533234
+      "required_token_positions_per_second": 27856.742457392622,
+      "required_pages_per_second_upper_estimate": 0.34588751080115504
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4058997463593883,
-    "producer_p95_seconds": 0.1907772520007711,
-    "gpu_update_seconds": 1.0043208479903556,
-    "producer_batches_per_second": 2.463662539750859,
-    "gpu_batches_per_second": 0.9956977414150053,
-    "producer_headroom": 2.4743076510846556,
+    "producer_batch_seconds": 0.07094033175782499,
+    "producer_p95_seconds": 0.1819227480000336,
+    "gpu_update_seconds": 0.9925065733112227,
+    "producer_batches_per_second": 14.096353586472988,
+    "gpu_batches_per_second": 1.0075500020758326,
+    "producer_headroom": 13.990723594293671,
     "sustainable": true,
-    "buffer_seconds": 60.25925087942134,
+    "buffer_seconds": 60.54290097198459,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366286.81253284216,
-    "upload_bytes_per_second": 3868822.892782808,
+    "generation_bytes_per_second": 388878.86046958825,
+    "upload_bytes_per_second": 3927443.8644016795,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.59662324894856,
+    "estimated_full_shards_per_hour": 83.44435082021461,
     "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": [
@@ -388,6 +388,12 @@
       "bytes": 11658602,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 0.9783495700003186,
+      "bytes": 7342912,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -37,3 +37,4 @@
 {"step": 3700, "loss": 0.7664323039352894, "lr": 0.0003, "grad_norm": 0.18385045648896226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 541090179, "tokens_per_second": 148339.78041008094, "data_wait_s": 1.9932999748561997e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 3800, "loss": 0.797960476949811, "lr": 0.0003, "grad_norm": 0.17936518091947842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 555775141, "tokens_per_second": 149119.53054489655, "data_wait_s": 1.7926000509760343e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 3900, "loss": 0.8140288442373276, "lr": 0.0003, "grad_norm": 0.19455745530295762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 570497907, "tokens_per_second": 148998.336972587, "data_wait_s": 1.1072000233980361e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 4000, "loss": 0.781925618648529, "lr": 0.0003, "grad_norm": 0.19247554432262717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 585075237, "tokens_per_second": 147804.51803401762, "data_wait_s": 9.609000699128956e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:641c71c5632619ce356681a67f0d75ded1d03aa9b0c83671ffc0cc610ba200eb
-size 81363
+oid sha256:19c2c2095daf8b61f99fdd3068135965b0eec109c713d2417603203d4070aa2a
+size 83446