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": 75942.0378351883,
-      "required_pages_per_second_upper_estimate": 0.005253846387243077
+      "required_token_positions_per_second": 76266.96378841305,
+      "required_pages_per_second_upper_estimate": 0.005276325518619249
     },
     {
       "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": 9492.754729398537,
-      "required_pages_per_second_upper_estimate": 0.00397441817393879
+      "required_token_positions_per_second": 9533.370473551631,
+      "required_pages_per_second_upper_estimate": 0.003991423137862505
     },
     {
       "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": 18985.509458797074,
-      "required_pages_per_second_upper_estimate": 0.003954180938713626
+      "required_token_positions_per_second": 19066.740947103262,
+      "required_pages_per_second_upper_estimate": 0.0039710993155094
     },
     {
       "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": 28478.26418819561,
-      "required_pages_per_second_upper_estimate": 0.0017072284542907048
+      "required_token_positions_per_second": 28600.111420654895,
+      "required_pages_per_second_upper_estimate": 0.0017145330098266875
     },
     {
       "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": 28478.26418819561,
-      "required_pages_per_second_upper_estimate": 0.0022140597779862093
+      "required_token_positions_per_second": 28600.111420654895,
+      "required_pages_per_second_upper_estimate": 0.0022235328643605843
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28478.26418819561,
-      "required_pages_per_second_upper_estimate": 0.3652885954284272
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28600.111420654895,
+      "required_pages_per_second_upper_estimate": 0.36685152089704975
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10996972166398677,
-    "producer_p95_seconds": 0.34619542799919145,
-    "gpu_update_seconds": 0.9708456883920699,
-    "producer_batches_per_second": 9.093412121706615,
-    "gpu_batches_per_second": 1.0300298100475844,
-    "producer_headroom": 8.828299951131052,
+    "producer_batch_seconds": 0.10181471108580809,
+    "producer_p95_seconds": 0.316304255000432,
+    "gpu_update_seconds": 0.9667095205801441,
+    "producer_batches_per_second": 9.821763371279552,
+    "gpu_batches_per_second": 1.034436900341974,
+    "producer_headroom": 9.494792159901277,
     "sustainable": true,
-    "buffer_seconds": 60.19243268030833,
+    "buffer_seconds": 60.90269979654908,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 158,
+    "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 581088.5892554799,
-    "upload_bytes_per_second": 5771072.048088812,
+    "generation_bytes_per_second": 563702.8394291286,
+    "upload_bytes_per_second": 5731589.848918298,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.34404210209034,
+    "estimated_full_shards_per_hour": 60.47875350549408,
     "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": 1.083468554999854,
-      "bytes": 6825645,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.0758820400005789,
       "bytes": 6474447,
@@ -562,6 +556,12 @@
       "bytes": 12351474,
       "pages": 36,
       "shards": 5
+    },
+    {
+      "seconds": 1.4946348649973515,
+      "bytes": 6250315,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -310,3 +310,4 @@
 {"step": 588100, "loss": 0.5901328325271606, "lr": 0.0003, "grad_norm": 0.16871667434024037, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86079988934, "tokens_per_second": 150928.4543544021, "data_wait_s": 1.1067000741604716e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 588200, "loss": 0.5681907534599304, "lr": 0.0003, "grad_norm": 0.17472111764928752, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86094666912, "tokens_per_second": 152947.53144042415, "data_wait_s": 1.0006999218603596e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 588300, "loss": 0.6095906496047974, "lr": 0.0003, "grad_norm": 0.15701132721614816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86109417988, "tokens_per_second": 152014.81540864406, "data_wait_s": 1.1331001587677747e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 588400, "loss": 0.5753762125968933, "lr": 0.0003, "grad_norm": 0.17385487216208018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86124218995, "tokens_per_second": 153909.21463850315, "data_wait_s": 1.2128999514970928e-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:a4ae39a2f7901d6c180aaeca9579d1619f8b71bdd61df8c5262b1c3a7bb278d1
-size 665679
+oid sha256:af17b80ecef6bc0dcac3ebc111285587acddea90feb58dfc0c9b29809a6a0f6d
+size 667797