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": 77079.70016643692,
-      "required_pages_per_second_upper_estimate": 0.005332552507059151
+      "required_token_positions_per_second": 77125.89896289956,
+      "required_pages_per_second_upper_estimate": 0.005335748646994413
     },
     {
       "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": 9634.962520804615,
-      "required_pages_per_second_upper_estimate": 0.004033957606564141
+      "required_token_positions_per_second": 9640.737370362445,
+      "required_pages_per_second_upper_estimate": 0.004036375415481433
     },
     {
       "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": 19269.92504160923,
-      "required_pages_per_second_upper_estimate": 0.004013417204070089
+      "required_token_positions_per_second": 19281.47474072489,
+      "required_pages_per_second_upper_estimate": 0.004015822701809833
     },
     {
       "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": 28904.887562413845,
-      "required_pages_per_second_upper_estimate": 0.0017328038741589136
+      "required_token_positions_per_second": 28922.212111087338,
+      "required_pages_per_second_upper_estimate": 0.0017338424543988354
     },
     {
       "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": 28904.887562413845,
-      "required_pages_per_second_upper_estimate": 0.0022472278688137706
+      "required_token_positions_per_second": 28922.212111087338,
+      "required_pages_per_second_upper_estimate": 0.002248574776270498
     },
     {
       "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": 28904.887562413845,
-      "required_pages_per_second_upper_estimate": 0.3707608620004085
+      "required_token_positions_per_second": 28922.212111087338,
+      "required_pages_per_second_upper_estimate": 0.37098308270914093
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.48172636720300943,
-    "producer_p95_seconds": 0.2378701830002683,
-    "gpu_update_seconds": 0.9565164348174727,
-    "producer_batches_per_second": 2.075867272547652,
-    "gpu_batches_per_second": 1.0454603429692508,
-    "producer_headroom": 1.9856011626915513,
+    "producer_batch_seconds": 0.07811161516397647,
+    "producer_p95_seconds": 0.2010691890009184,
+    "gpu_update_seconds": 0.955943476723246,
+    "producer_batches_per_second": 12.8021933473113,
+    "gpu_batches_per_second": 1.0460869542493973,
+    "producer_headroom": 12.238173218111974,
     "sustainable": true,
-    "buffer_seconds": 60.26053539350078,
+    "buffer_seconds": 60.224439033564494,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 463021.6283930681,
-    "upload_bytes_per_second": 2144968.281865776,
+    "generation_bytes_per_second": 463522.1329120884,
+    "upload_bytes_per_second": 2189594.465258072,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.35366286129029,
+    "estimated_full_shards_per_hour": 99.46105947992315,
     "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": 0.968915202000062,
-      "bytes": 6356080,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.188576070999261,
       "bytes": 5752447,
@@ -562,6 +556,12 @@
       "bytes": 1284941,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.8045908780004538,
+      "bytes": 17039792,
+      "pages": 37,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -154,3 +154,4 @@
 {"step": 660400, "loss": 0.5669741630554199, "lr": 0.0001, "grad_norm": 0.20515818367769723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96665258997, "tokens_per_second": 153529.86450996198, "data_wait_s": 1.4510000255540945e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 660500, "loss": 0.5582073926925659, "lr": 0.0001, "grad_norm": 0.1895254287130386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96679880377, "tokens_per_second": 152766.77688810934, "data_wait_s": 1.085700023395475e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 660600, "loss": 0.522675633430481, "lr": 0.0001, "grad_norm": 0.19461194593324185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96694521168, "tokens_per_second": 152697.3829239485, "data_wait_s": 1.2554000932141207e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 660700, "loss": 0.5982367396354675, "lr": 0.0001, "grad_norm": 0.20127042855920593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96709222604, "tokens_per_second": 153484.15191274587, "data_wait_s": 8.015000275918283e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:367cf48cb43a2a5e9565110021641cd2014fad13902f52802645ddd1021fab05
-size 332922
+oid sha256:9483cc51a064f8abec6b9e8acf9075c6f329cd8444ac8ed4a2c91eae76c31bd7
+size 335040