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": 76810.17355169753,
-      "required_pages_per_second_upper_estimate": 0.005313906030463572
+      "required_token_positions_per_second": 75445.98926746365,
+      "required_pages_per_second_upper_estimate": 0.005219528596336634
     },
     {
       "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": 9601.271693962191,
-      "required_pages_per_second_upper_estimate": 0.004019851960909686
+      "required_token_positions_per_second": 9430.748658432956,
+      "required_pages_per_second_upper_estimate": 0.003948457526859503
     },
     {
       "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": 19202.543387924383,
-      "required_pages_per_second_upper_estimate": 0.00399938338258124
+      "required_token_positions_per_second": 18861.497316865913,
+      "required_pages_per_second_upper_estimate": 0.003928352479969474
     },
     {
       "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": 28803.815081886572,
-      "required_pages_per_second_upper_estimate": 0.0017267447332800455
+      "required_token_positions_per_second": 28292.245975298865,
+      "required_pages_per_second_upper_estimate": 0.0016960769464609098
     },
     {
       "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": 28803.815081886572,
-      "required_pages_per_second_upper_estimate": 0.002239369928023633
+      "required_token_positions_per_second": 28292.245975298865,
+      "required_pages_per_second_upper_estimate": 0.002199597680141135
     },
     {
       "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": 28803.815081886572,
-      "required_pages_per_second_upper_estimate": 0.36946441274337904
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28292.245975298865,
+      "required_pages_per_second_upper_estimate": 0.36290255352418344
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.047403146781292094,
-    "producer_p95_seconds": 0.1627522709995901,
-    "gpu_update_seconds": 0.9598728474474406,
-    "producer_batches_per_second": 21.095645920170334,
-    "gpu_batches_per_second": 1.0418046542927726,
-    "producer_headroom": 20.249137718136883,
+    "producer_batch_seconds": 0.5701434108203856,
+    "producer_p95_seconds": 0.18239391499992053,
+    "gpu_update_seconds": 0.977228885403395,
+    "producer_batches_per_second": 1.7539446760615705,
+    "gpu_batches_per_second": 1.0233017207501036,
+    "producer_headroom": 1.7140054008468673,
     "sustainable": true,
-    "buffer_seconds": 60.47198938918876,
+    "buffer_seconds": 60.58819089501049,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 426391.49805789656,
-    "upload_bytes_per_second": 5635024.902192756,
+    "generation_bytes_per_second": 372502.4345470598,
+    "upload_bytes_per_second": 5729523.699225536,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.49368959715531,
+    "estimated_full_shards_per_hour": 79.9303510409245,
     "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": [
@@ -430,6 +430,12 @@
       "bytes": 6290854,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1961185629998,
+      "bytes": 11182870,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -48,3 +48,4 @@
 {"step": 561900, "loss": 0.5914993286132812, "lr": 0.0003, "grad_norm": 0.14266866861507754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82244640620, "tokens_per_second": 153630.32665940447, "data_wait_s": 1.4713000382471364e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 562000, "loss": 0.5662038326263428, "lr": 0.0003, "grad_norm": 0.1866444866498162, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82259344566, "tokens_per_second": 153271.57740873244, "data_wait_s": 1.495099968451541e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 562100, "loss": 0.5918933153152466, "lr": 0.0003, "grad_norm": 0.16625810725396656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82273953443, "tokens_per_second": 152768.37790005782, "data_wait_s": 1.5190999874903355e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 562200, "loss": 0.5662493109703064, "lr": 0.0003, "grad_norm": 0.14734996078220597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82288728817, "tokens_per_second": 152798.27289510635, "data_wait_s": 9.854999916569795e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8b62dee51b7c2335c16f86fccac5c6bf125180b629cd7d9e68c089fef855d6d1
-size 106848
+oid sha256:8506d9a51f8b8943164626837615bb54672be58471744df26df23938cdce5b3b
+size 108966