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": 76301.72529565268,
-      "required_pages_per_second_upper_estimate": 0.005278730400347896
+      "required_token_positions_per_second": 76181.77984223538,
+      "required_pages_per_second_upper_estimate": 0.005270432295568691
     },
     {
       "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": 9537.715661956585,
-      "required_pages_per_second_upper_estimate": 0.003993242377509807
+      "required_token_positions_per_second": 9522.722480279423,
+      "required_pages_per_second_upper_estimate": 0.003986965045434816
     },
     {
       "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": 19075.43132391317,
-      "required_pages_per_second_upper_estimate": 0.0039729092918182555
+      "required_token_positions_per_second": 19045.444960558845,
+      "required_pages_per_second_upper_estimate": 0.003966663923125131
     },
     {
       "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": 28613.146985869756,
-      "required_pages_per_second_upper_estimate": 0.0017153144720571628
+      "required_token_positions_per_second": 28568.167440838268,
+      "required_pages_per_second_upper_estimate": 0.0017126180170123136
     },
     {
       "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": 28613.146985869756,
-      "required_pages_per_second_upper_estimate": 0.0022245463222186476
+      "required_token_positions_per_second": 28568.167440838268,
+      "required_pages_per_second_upper_estimate": 0.002221049360436559
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28613.146985869756,
-      "required_pages_per_second_upper_estimate": 0.36701872713112654
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28568.167440838268,
+      "required_pages_per_second_upper_estimate": 0.3664417778227353
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3695861882266058,
-    "producer_p95_seconds": 0.3610797579967766,
-    "gpu_update_seconds": 0.966269107471947,
-    "producer_batches_per_second": 2.705728817406093,
-    "gpu_batches_per_second": 1.0349083834588308,
-    "producer_headroom": 2.6144621694561128,
+    "producer_batch_seconds": 0.08544708759364994,
+    "producer_p95_seconds": 0.29981768099969486,
+    "gpu_update_seconds": 0.9677904631879577,
+    "producer_batches_per_second": 11.703149026630086,
+    "gpu_batches_per_second": 1.0332815191275415,
+    "producer_headroom": 11.326196017240028,
     "sustainable": true,
-    "buffer_seconds": 60.87495377073267,
+    "buffer_seconds": 60.00300871765338,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 42223.51039486148,
-    "upload_bytes_per_second": 5086345.062478578,
+    "generation_bytes_per_second": 42249.4847962826,
+    "upload_bytes_per_second": 5221475.167813206,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 9.060182417720634,
+    "estimated_full_shards_per_hour": 9.065755919612489,
     "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": 1.0994338210002752,
-      "bytes": 2592435,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.061899332999019,
       "bytes": 6323038,
@@ -562,6 +556,12 @@
       "bytes": 4766956,
       "pages": 1,
       "shards": 1
+    },
+    {
+      "seconds": 1.2835849650000455,
+      "bytes": 14052046,
+      "pages": 50,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -325,3 +325,4 @@
 {"step": 717500, "loss": 0.5696074962615967, "lr": 8e-05, "grad_norm": 0.21744040136766699, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105022269675, "tokens_per_second": 152364.1583252598, "data_wait_s": 1.673299993854016e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 717600, "loss": 0.5513715147972107, "lr": 8e-05, "grad_norm": 0.21986178507525256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105036984974, "tokens_per_second": 6605.745511029615, "data_wait_s": 1.2858996342401952e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 717700, "loss": 0.5750818848609924, "lr": 8e-05, "grad_norm": 0.21498823118512472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105051626347, "tokens_per_second": 152844.6147458412, "data_wait_s": 9.174000297207385e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 717800, "loss": 0.6104263663291931, "lr": 8e-05, "grad_norm": 0.27834772034823047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105066320787, "tokens_per_second": 152397.20697045236, "data_wait_s": 1.7195001419167966e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:38c786a100eda0757e495bdc87c2452897964231800b41cef674b29308041014
-size 697449
+oid sha256:1be258fcb7230c88a0446595a16ffa37f6d1574bac37c295eaaaa0a80a60b082
+size 699567