Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74901.23298382605,
-      "required_pages_per_second_upper_estimate": 0.006492249039598604
+      "required_token_positions_per_second": 74724.08890194965,
+      "required_pages_per_second_upper_estimate": 0.006476894639549129
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9362.654122978256,
-      "required_pages_per_second_upper_estimate": 0.00373490271380974
+      "required_token_positions_per_second": 9340.511112743707,
+      "required_pages_per_second_upper_estimate": 0.0037260695359596723
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18725.308245956512,
-      "required_pages_per_second_upper_estimate": 0.004331770508532583
+      "required_token_positions_per_second": 18681.022225487413,
+      "required_pages_per_second_upper_estimate": 0.004321525717104397
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28087.962368934768,
-      "required_pages_per_second_upper_estimate": 0.0017104095421186338
+      "required_token_positions_per_second": 28021.53333823112,
+      "required_pages_per_second_upper_estimate": 0.00170636436267497
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28087.962368934768,
-      "required_pages_per_second_upper_estimate": 0.0021622890956168143
+      "required_token_positions_per_second": 28021.53333823112,
+      "required_pages_per_second_upper_estimate": 0.002157175204945921
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28087.962368934768,
-      "required_pages_per_second_upper_estimate": 0.34875848825924444
+      "required_token_positions_per_second": 28021.53333823112,
+      "required_pages_per_second_upper_estimate": 0.34793366202156917
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.28280490235920297,
-    "producer_p95_seconds": 0.3295628800005943,
-    "gpu_update_seconds": 0.9843362660788322,
-    "producer_batches_per_second": 3.5360065955640896,
-    "gpu_batches_per_second": 1.0159129907745503,
-    "producer_headroom": 3.480619529107679,
+    "producer_batch_seconds": 0.11200499768705185,
+    "producer_p95_seconds": 0.3322004139990895,
+    "gpu_update_seconds": 0.9866697752145672,
+    "producer_batches_per_second": 8.928173033796718,
+    "gpu_batches_per_second": 1.0135103203931972,
+    "producer_headroom": 8.809158480332966,
     "sustainable": true,
-    "buffer_seconds": 60.04451223080876,
+    "buffer_seconds": 60.1868562880886,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 411121.69498331135,
-    "upload_bytes_per_second": 2865090.672524844,
+    "generation_bytes_per_second": 409745.259217941,
+    "upload_bytes_per_second": 2878932.466301815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.21714532017236,
+    "estimated_full_shards_per_hour": 87.92179424670861,
     "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": 1.235119819000829,
-      "bytes": 5680772,
-      "pages": 24,
-      "shards": 2
-    },
     {
       "seconds": 1.104966183000215,
       "bytes": 6761570,
@@ -562,6 +556,12 @@
       "bytes": 11885196,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.9526699870002631,
+      "bytes": 6951757,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 219200, "loss": 0.5834460780024529, "lr": 0.0003, "grad_norm": 0.15104916600150237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32083818204, "tokens_per_second": 149474.60381156084, "data_wait_s": 1.1145999451400712e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 219300, "loss": 0.5816761590540409, "lr": 0.0003, "grad_norm": 0.14476674569779652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32098461872, "tokens_per_second": 148982.575581119, "data_wait_s": 1.609200262464583e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 219400, "loss": 0.5835527759045362, "lr": 0.0003, "grad_norm": 0.15944790522468322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32113115322, "tokens_per_second": 148725.52193367132, "data_wait_s": 1.5837998944334686e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 219500, "loss": 0.5853289980441332, "lr": 0.0003, "grad_norm": 0.1496752562692594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32127750282, "tokens_per_second": 148218.00638480662, "data_wait_s": 1.4034998457645997e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f4b7a3813a5fee88225eb7d1c9f3440bffb51e888070ffbb677699acd57d4b4c
-size 520089
+oid sha256:4422b17dd9467e83ca7c4ba5b8bdc098bac745227faec10bf99930b02bf6567d
+size 522207