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": 74737.58671276536,
-      "required_pages_per_second_upper_estimate": 0.006478064595580746
+      "required_token_positions_per_second": 75967.22746347329,
+      "required_pages_per_second_upper_estimate": 0.006584646739356119
     },
     {
       "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": 9342.19833909567,
-      "required_pages_per_second_upper_estimate": 0.0037267425957777526
+      "required_token_positions_per_second": 9495.903432934161,
+      "required_pages_per_second_upper_estimate": 0.0037880578558058724
     },
     {
       "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": 18684.39667819134,
-      "required_pages_per_second_upper_estimate": 0.004322306337349111
+      "required_token_positions_per_second": 18991.806865868322,
+      "required_pages_per_second_upper_estimate": 0.004393420274033661
     },
     {
       "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": 28026.59501728701,
-      "required_pages_per_second_upper_estimate": 0.0017066725923728962
+      "required_token_positions_per_second": 28487.71029880248,
+      "required_pages_per_second_upper_estimate": 0.0017347520937322784
     },
     {
       "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": 28026.59501728701,
-      "required_pages_per_second_upper_estimate": 0.002157564867011268
+      "required_token_positions_per_second": 28487.71029880248,
+      "required_pages_per_second_upper_estimate": 0.0021930627978311245
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28026.59501728701,
-      "required_pages_per_second_upper_estimate": 0.34799651113509333
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28487.71029880248,
+      "required_pages_per_second_upper_estimate": 0.3537220196779428
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.8038231212968725,
-    "producer_p95_seconds": 0.218273993999901,
-    "gpu_update_seconds": 0.98649157997775,
-    "producer_batches_per_second": 1.2440547845732772,
-    "gpu_batches_per_second": 1.0136933961692352,
-    "producer_headroom": 1.2272495700125716,
+    "producer_batch_seconds": 0.07762978306240598,
+    "producer_p95_seconds": 0.1995838780003396,
+    "gpu_update_seconds": 0.9705237700750635,
+    "producer_batches_per_second": 12.881653929086827,
+    "gpu_batches_per_second": 1.0303714662471963,
+    "producer_headroom": 12.501951336059603,
     "sustainable": true,
-    "buffer_seconds": 60.175986378642754,
+    "buffer_seconds": 60.172473744653935,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 199529.231291924,
-    "upload_bytes_per_second": 6549249.712682356,
+    "generation_bytes_per_second": 191797.30140295514,
+    "upload_bytes_per_second": 6593767.810258815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 42.814328232462785,
+    "estimated_full_shards_per_hour": 41.15523606840602,
     "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": 1.0683630309999899,
-      "bytes": 9860015,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 0.8969114139999874,
       "bytes": 7503017,
@@ -562,6 +556,12 @@
       "bytes": 2046300,
       "pages": 19,
       "shards": 2
+    },
+    {
+      "seconds": 1.6571312910000415,
+      "bytes": 16761254,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -96,3 +96,4 @@
 {"step": 9600, "loss": 0.7038426920771599, "lr": 0.0003, "grad_norm": 0.157569396104347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1403737264, "tokens_per_second": 146875.71258061347, "data_wait_s": 1.1778000043705106e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 9700, "loss": 0.7019376754760742, "lr": 0.0003, "grad_norm": 0.15032665239691817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1418274502, "tokens_per_second": 148051.27805517896, "data_wait_s": 1.4580999049940147e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 9800, "loss": 0.6842157915234566, "lr": 0.0003, "grad_norm": 0.14682688827643217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1432900392, "tokens_per_second": 92317.45613918074, "data_wait_s": 1.2158001482021064e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 9900, "loss": 0.801698736846447, "lr": 0.0003, "grad_norm": 0.1893745328406307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1447622497, "tokens_per_second": 151120.41772944975, "data_wait_s": 8.74599936651066e-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 sha256:0a26fa882aa2bce20000897bd31fb2f2567ad30a55e109372830a5a587b29bfd
-size 205033
+oid sha256:6b642558b264e1fdf879891ddee550aae4c771ae0ea1eed27abfb269bbea9b23
+size 207116