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": 76398.66995718279,
-      "required_pages_per_second_upper_estimate": 0.005285437256975154
+      "required_token_positions_per_second": 75473.26619311613,
+      "required_pages_per_second_upper_estimate": 0.005221415677344459
     },
     {
       "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": 9549.833744647849,
-      "required_pages_per_second_upper_estimate": 0.003998315965678297
+      "required_token_positions_per_second": 9434.158274139516,
+      "required_pages_per_second_upper_estimate": 0.003949885061754967
     },
     {
       "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": 19099.667489295698,
-      "required_pages_per_second_upper_estimate": 0.003977957045916775
+      "required_token_positions_per_second": 18868.316548279032,
+      "required_pages_per_second_upper_estimate": 0.003929772746037601
     },
     {
       "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": 28649.50123394355,
-      "required_pages_per_second_upper_estimate": 0.0017174938537194639
+      "required_token_positions_per_second": 28302.474822418546,
+      "required_pages_per_second_upper_estimate": 0.0016966901502271895
     },
     {
       "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": 28649.50123394355,
-      "required_pages_per_second_upper_estimate": 0.0022273727051009473
+      "required_token_positions_per_second": 28302.474822418546,
+      "required_pages_per_second_upper_estimate": 0.0022003929280127467
     },
     {
       "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": 28649.50123394355,
-      "required_pages_per_second_upper_estimate": 0.3674850403912668
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28302.474822418546,
+      "required_pages_per_second_upper_estimate": 0.36303375819215433
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0936250037343882,
-    "producer_p95_seconds": 0.2512930659977428,
-    "gpu_update_seconds": 0.9650429783832682,
-    "producer_batches_per_second": 10.680907451143874,
-    "gpu_batches_per_second": 1.036223279584185,
-    "producer_headroom": 10.307534738487927,
+    "producer_batch_seconds": 0.4269218717733736,
+    "producer_p95_seconds": 0.22597827299978235,
+    "gpu_update_seconds": 0.9768757033960813,
+    "producer_batches_per_second": 2.342348954496382,
+    "gpu_batches_per_second": 1.0236716877321523,
+    "producer_headroom": 2.2881837825227285,
     "sustainable": true,
-    "buffer_seconds": 60.7977076381459,
+    "buffer_seconds": 60.56629361055704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364412.08012096834,
-    "upload_bytes_per_second": 5347191.294908707,
+    "generation_bytes_per_second": 328530.78726124176,
+    "upload_bytes_per_second": 5345252.410769912,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.19434931489742,
+    "estimated_full_shards_per_hour": 70.49505914094867,
     "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": 1.2992984630000137,
-      "bytes": 10017513,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 4.1799396490005165,
       "bytes": 5932760,
@@ -562,6 +556,12 @@
       "bytes": 7245209,
       "pages": 27,
       "shards": 3
+    },
+    {
+      "seconds": 1.4228277920010441,
+      "bytes": 10524529,
+      "pages": 42,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -236,3 +236,4 @@
 {"step": 580700, "loss": 0.5879694819450378, "lr": 0.0003, "grad_norm": 0.17708666210033602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84997580565, "tokens_per_second": 152314.54234724957, "data_wait_s": 1.52689972310327e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 580800, "loss": 0.5479899644851685, "lr": 0.0003, "grad_norm": 0.1745376355488399, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85012037807, "tokens_per_second": 149893.2469305483, "data_wait_s": 1.4523000572808087e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 580900, "loss": 0.5429603457450867, "lr": 0.0003, "grad_norm": 0.14829949338349663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85026753544, "tokens_per_second": 153268.9895131118, "data_wait_s": 8.085000445134938e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 581000, "loss": 0.6548506617546082, "lr": 0.0003, "grad_norm": 0.1879257302904096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85041352123, "tokens_per_second": 129773.14262327352, "data_wait_s": 1.098900247598067e-05, "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 sha256:828a7cb7ea3d73987af91d6510900aee8e26dd1935589628f17f1f642b2c9b50
-size 508164
+oid sha256:87979546df02b6fe4b1866e6170dc02a203cc8a96d655b63e0cd62f6d78b7b11
+size 510282