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": 76295.37193780951,
-      "required_pages_per_second_upper_estimate": 0.0066130894707170165
+      "required_token_positions_per_second": 75726.36233090289,
+      "required_pages_per_second_upper_estimate": 0.006563769160132014
     },
     {
       "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": 9536.921492226189,
-      "required_pages_per_second_upper_estimate": 0.0038044205729321004
+      "required_token_positions_per_second": 9465.795291362861,
+      "required_pages_per_second_upper_estimate": 0.003776047267976249
     },
     {
       "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": 19073.842984452378,
-      "required_pages_per_second_upper_estimate": 0.004412397886281706
+      "required_token_positions_per_second": 18931.590582725723,
+      "required_pages_per_second_upper_estimate": 0.004379490296699008
     },
     {
       "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": 28610.764476678567,
-      "required_pages_per_second_upper_estimate": 0.0017422454475495566
+      "required_token_positions_per_second": 28397.385874088584,
+      "required_pages_per_second_upper_estimate": 0.001729251810162834
     },
     {
       "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": 28610.764476678567,
-      "required_pages_per_second_upper_estimate": 0.0022025358490797915
+      "required_token_positions_per_second": 28397.385874088584,
+      "required_pages_per_second_upper_estimate": 0.0021861093735826382
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28610.764476678567,
-      "required_pages_per_second_upper_estimate": 0.3552499407313231
+      "required_token_positions_per_second": 28397.385874088584,
+      "required_pages_per_second_upper_estimate": 0.3526004926193996
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11577524164027864,
-    "producer_p95_seconds": 0.352708274993347,
-    "gpu_update_seconds": 0.9663495717682293,
-    "producer_batches_per_second": 8.63742528913968,
-    "gpu_batches_per_second": 1.0348222105280152,
-    "producer_headroom": 8.346772229340203,
+    "producer_batch_seconds": 0.17491744328924597,
+    "producer_p95_seconds": 0.3678696260030847,
+    "gpu_update_seconds": 0.9736107444040345,
+    "producer_batches_per_second": 5.716982715934087,
+    "gpu_batches_per_second": 1.0271045238023937,
+    "producer_headroom": 5.566115797805585,
     "sustainable": true,
-    "buffer_seconds": 60.88002302139845,
+    "buffer_seconds": 60.36386615305014,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399339.8377125002,
-    "upload_bytes_per_second": 5697066.613027273,
+    "generation_bytes_per_second": 409176.53945980646,
+    "upload_bytes_per_second": 5677622.867585142,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.68903301745657,
+    "estimated_full_shards_per_hour": 87.79976022573133,
     "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": 0.9628497100020468,
-      "bytes": 5747199,
-      "pages": 36,
-      "shards": 2
-    },
-    {
-      "seconds": 0.9260915579980065,
-      "bytes": 968750,
-      "pages": 30,
-      "shards": 1
-    },
     {
       "seconds": 1.5047391669977515,
       "bytes": 11683531,
@@ -562,6 +550,18 @@
       "bytes": 12230231,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.34678302899556,
+      "bytes": 6766636,
+      "pages": 30,
+      "shards": 2
+    },
+    {
+      "seconds": 0.8585810450022109,
+      "bytes": 306974,
+      "pages": 9,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -296,3 +296,4 @@
 {"step": 309800, "loss": 0.5477604027837515, "lr": 0.0003, "grad_norm": 0.15359239092624632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45346368767, "tokens_per_second": 148047.64556304982, "data_wait_s": 1.1169002391397953e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 309900, "loss": 0.6197275631129742, "lr": 0.0003, "grad_norm": 0.16832917274897194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45360906797, "tokens_per_second": 149268.5298529645, "data_wait_s": 1.0035000741481781e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 310000, "loss": 0.5801891293376684, "lr": 0.0003, "grad_norm": 0.14141978433655597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45375546310, "tokens_per_second": 150883.49656436348, "data_wait_s": 1.284800237044692e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 310100, "loss": 0.6318192221224308, "lr": 0.0003, "grad_norm": 0.19374741506262244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45390295004, "tokens_per_second": 136630.82280010704, "data_wait_s": 9.717005013953894e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7f3222cd70373b9db862598f3f2a2c7d4bc93542219d0324b0343b2fa4e3065
-size 636027
+oid sha256:395a91ba8dfe1b109c8b1ce88b3bee84fa5563e60f8e423c5ce2b5bc2c20dc09
+size 638145