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": 76883.19428704698,
-      "required_pages_per_second_upper_estimate": 0.005318957774366496
+      "required_token_positions_per_second": 75972.6094058172,
+      "required_pages_per_second_upper_estimate": 0.005255961399435002
     },
     {
       "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": 9610.399285880872,
-      "required_pages_per_second_upper_estimate": 0.004023673493040243
+      "required_token_positions_per_second": 9496.57617572715,
+      "required_pages_per_second_upper_estimate": 0.0039760181337157065
     },
     {
       "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": 19220.798571761745,
-      "required_pages_per_second_upper_estimate": 0.00400318545595299
+      "required_token_positions_per_second": 18993.1523514543,
+      "required_pages_per_second_upper_estimate": 0.003955772751697493
     },
     {
       "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": 28831.19785764262,
-      "required_pages_per_second_upper_estimate": 0.0017283862888755428
+      "required_token_positions_per_second": 28489.728527181447,
+      "required_pages_per_second_upper_estimate": 0.0017079157239078735
     },
     {
       "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": 28831.19785764262,
-      "required_pages_per_second_upper_estimate": 0.002241498818394564
+      "required_token_positions_per_second": 28489.728527181447,
+      "required_pages_per_second_upper_estimate": 0.00221495107991606
     },
     {
       "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": 28831.19785764262,
-      "required_pages_per_second_upper_estimate": 0.3698156495894437
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28489.728527181447,
+      "required_pages_per_second_upper_estimate": 0.3654356476594893
     }
   ],
   "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": 0.2317672585859043,
-    "producer_p95_seconds": 0.3118988310015993,
-    "gpu_update_seconds": 0.9589611967048752,
-    "producer_batches_per_second": 4.314673289494646,
-    "gpu_batches_per_second": 1.0427950614020045,
-    "producer_headroom": 4.137604261084347,
+    "producer_batch_seconds": 0.13181334107025577,
+    "producer_p95_seconds": 0.3053428150014952,
+    "gpu_update_seconds": 0.9704550176257954,
+    "producer_batches_per_second": 7.586485494415968,
+    "gpu_batches_per_second": 1.0304444635120604,
+    "producer_headroom": 7.36234291420129,
     "sustainable": true,
-    "buffer_seconds": 60.41455539240714,
+    "buffer_seconds": 60.16821109279932,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 486917.8200112482,
-    "upload_bytes_per_second": 5638268.863349004,
+    "generation_bytes_per_second": 487308.35354792146,
+    "upload_bytes_per_second": 5712213.984826128,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.48122930768093,
+    "estimated_full_shards_per_hour": 104.56502871349556,
     "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": 1.2669715150004777,
-      "bytes": 6065248,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.0634998090026784,
       "bytes": 7147043,
@@ -562,6 +556,12 @@
       "bytes": 1013057,
       "pages": 28,
       "shards": 1
+    },
+    {
+      "seconds": 1.2470868550008163,
+      "bytes": 11490913,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -318,3 +318,4 @@
 {"step": 588900, "loss": 0.5647330284118652, "lr": 0.0003, "grad_norm": 0.1624249155753471, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86197233219, "tokens_per_second": 152589.67730645378, "data_wait_s": 1.138900188379921e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 589000, "loss": 0.5505220890045166, "lr": 0.0003, "grad_norm": 0.15693049434588788, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86211931171, "tokens_per_second": 152834.818940623, "data_wait_s": 1.4011999155627564e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 589100, "loss": 0.5802993774414062, "lr": 0.0003, "grad_norm": 0.1582367300546252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86226618796, "tokens_per_second": 153113.84424693335, "data_wait_s": 1.2295000487938523e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 589200, "loss": 0.6048529744148254, "lr": 0.0003, "grad_norm": 0.16448242669712304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86241351148, "tokens_per_second": 153462.7825227095, "data_wait_s": 9.401999705005437e-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:c9ed46627f303e031696e99b01e4022a0e03ae1b0370c4446528afc17edd5b25
-size 682623
+oid sha256:0bdb0fe4c1a3ca0fadcaa18eb64ae41d7652aaca1b57ea45f41eaf4292a5d8da
+size 684741