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": 77500.51802138316,
-      "required_pages_per_second_upper_estimate": 0.005361665662696281
+      "required_token_positions_per_second": 77489.62290672486,
+      "required_pages_per_second_upper_estimate": 0.005360911913384072
     },
     {
       "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": 9687.564752672895,
-      "required_pages_per_second_upper_estimate": 0.004055981062587879
+      "required_token_positions_per_second": 9686.202863340608,
+      "required_pages_per_second_upper_estimate": 0.004055410867963933
     },
     {
       "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": 19375.12950534579,
-      "required_pages_per_second_upper_estimate": 0.004035328519438134
+      "required_token_positions_per_second": 19372.405726681216,
+      "required_pages_per_second_upper_estimate": 0.004034761228173177
     },
     {
       "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": 29062.694258018688,
-      "required_pages_per_second_upper_estimate": 0.0017422641446035517
+      "required_token_positions_per_second": 29058.608590021824,
+      "required_pages_per_second_upper_estimate": 0.0017420192150455936
     },
     {
       "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": 29062.694258018688,
-      "required_pages_per_second_upper_estimate": 0.00225949664527875
+      "required_token_positions_per_second": 29058.608590021824,
+      "required_pages_per_second_upper_estimate": 0.0022591790025629477
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 29062.694258018688,
-      "required_pages_per_second_upper_estimate": 0.37278503685199893
+      "required_token_positions_per_second": 29058.608590021824,
+      "required_pages_per_second_upper_estimate": 0.37273263028978365
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11004324976548219,
-    "producer_p95_seconds": 0.22295468800075469,
-    "gpu_update_seconds": 0.9513226734775851,
-    "producer_batches_per_second": 9.087336134939147,
-    "gpu_batches_per_second": 1.051168050420236,
-    "producer_headroom": 8.644988906679774,
+    "producer_batch_seconds": 0.07799237862496966,
+    "producer_p95_seconds": 0.21192387700284598,
+    "gpu_update_seconds": 0.9514564303499996,
+    "producer_batches_per_second": 12.821765634416039,
+    "gpu_batches_per_second": 1.0510202759701182,
+    "producer_headroom": 12.199351361305782,
     "sustainable": true,
-    "buffer_seconds": 60.88465110256545,
+    "buffer_seconds": 60.893211542399975,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 250929.12900663287,
-    "upload_bytes_per_second": 5221343.548163529,
+    "generation_bytes_per_second": 240161.06641518546,
+    "upload_bytes_per_second": 5267032.726928931,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.84354975365867,
+    "estimated_full_shards_per_hour": 51.53297418920205,
     "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": 0.9223496029990201,
-      "bytes": 897417,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.0840730929994606,
       "bytes": 6199927,
@@ -562,6 +556,12 @@
       "bytes": 8159623,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.2689961300020514,
+      "bytes": 6434173,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -193,3 +193,4 @@
 {"step": 576400, "loss": 0.6187487840652466, "lr": 0.0003, "grad_norm": 0.16270405619755987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84367598981, "tokens_per_second": 151528.78434761916, "data_wait_s": 9.620001947041601e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 576500, "loss": 0.6035013794898987, "lr": 0.0003, "grad_norm": 0.16464012626293767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84382130335, "tokens_per_second": 151303.1980639094, "data_wait_s": 1.166500078397803e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 576600, "loss": 0.5461407899856567, "lr": 0.0003, "grad_norm": 0.15148935357515936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84396819250, "tokens_per_second": 153281.5926600468, "data_wait_s": 9.707997378427535e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 576700, "loss": 0.6300444006919861, "lr": 0.0003, "grad_norm": 0.1907281988405383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84411542521, "tokens_per_second": 153690.82856286562, "data_wait_s": 7.604998245369643e-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 sha256:79cd24a821c536fb9fee1897afc9ea9f03a20bccf49f63e4c38125afc8e0e502
-size 416307
+oid sha256:4199db6a18c3f4735f29c76418cd55075299da652fcd70d40eba6b274be4a0f4
+size 418425