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": 77165.94041731741,
-      "required_pages_per_second_upper_estimate": 0.005338518807720015
+      "required_token_positions_per_second": 75591.9889616019,
+      "required_pages_per_second_upper_estimate": 0.005229629193943082
     },
     {
       "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": 9645.742552164676,
-      "required_pages_per_second_upper_estimate": 0.004038470980581946
+      "required_token_positions_per_second": 9448.998620200238,
+      "required_pages_per_second_upper_estimate": 0.003956098404748926
     },
     {
       "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": 19291.485104329353,
-      "required_pages_per_second_upper_estimate": 0.004017907596557602
+      "required_token_positions_per_second": 18897.997240400477,
+      "required_pages_per_second_upper_estimate": 0.00393595445147401
     },
     {
       "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": 28937.22765649403,
-      "required_pages_per_second_upper_estimate": 0.0017347426134185562
+      "required_token_positions_per_second": 28346.995860600717,
+      "required_pages_per_second_upper_estimate": 0.0016993591184865215
     },
     {
       "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": 28937.22765649403,
-      "required_pages_per_second_upper_estimate": 0.002249742168879464
+      "required_token_positions_per_second": 28346.995860600717,
+      "required_pages_per_second_upper_estimate": 0.002203854242904059
     },
     {
       "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": 28937.22765649403,
-      "required_pages_per_second_upper_estimate": 0.37117568600318146
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28346.995860600717,
+      "required_pages_per_second_upper_estimate": 0.3636048262669888
     }
   ],
   "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": 0.11061520339842446,
-    "producer_p95_seconds": 0.25236058900190983,
-    "gpu_update_seconds": 0.9554474370593444,
-    "producer_batches_per_second": 9.040348607398062,
-    "gpu_batches_per_second": 1.0466300512331463,
-    "producer_headroom": 8.637577907061493,
+    "producer_batch_seconds": 0.499600050390427,
+    "producer_p95_seconds": 0.24169112399977166,
+    "gpu_update_seconds": 0.9753414483835218,
+    "producer_batches_per_second": 2.0016010791402463,
+    "gpu_batches_per_second": 1.0252819683376995,
+    "producer_headroom": 1.9522444956146678,
     "sustainable": true,
-    "buffer_seconds": 60.1931885347387,
+    "buffer_seconds": 60.47116979977835,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324896.87078706594,
-    "upload_bytes_per_second": 5407800.929262618,
+    "generation_bytes_per_second": 343661.1242827549,
+    "upload_bytes_per_second": 5426679.380613501,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.7153052588366,
+    "estimated_full_shards_per_hour": 73.74167724954592,
     "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": 1.0670534790006059,
-      "bytes": 5802765,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9111343770000531,
       "bytes": 2649643,
@@ -562,6 +556,12 @@
       "bytes": 1121895,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.125895612000022,
+      "bytes": 7606699,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -205,3 +205,4 @@
 {"step": 545500, "loss": 0.5636225938796997, "lr": 0.0003, "grad_norm": 0.177380601657661, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79843923087, "tokens_per_second": 153646.50909028458, "data_wait_s": 9.675997716840357e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 545600, "loss": 0.5654869079589844, "lr": 0.0003, "grad_norm": 0.15474211752709327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79858630896, "tokens_per_second": 153291.50754398707, "data_wait_s": 1.716499900794588e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 545700, "loss": 0.5606106519699097, "lr": 0.0003, "grad_norm": 0.1571157613446879, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79873456688, "tokens_per_second": 139604.44434697292, "data_wait_s": 1.5087000065250322e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 545800, "loss": 0.627779483795166, "lr": 0.0003, "grad_norm": 0.21038275067424334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79888209184, "tokens_per_second": 153975.8169430534, "data_wait_s": 1.6963000234682113e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:caa9b267f3ef704d7c1a15a4246f89fd5964e017ba29ff65c59f12c424835415
-size 441723
+oid sha256:0b104727a52d4573e0a1b800d046fc24249108929723603bb1e6475d0d13ccd6
+size 443841