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": 75695.05015589023,
-      "required_pages_per_second_upper_estimate": 0.006561055100161899
+      "required_token_positions_per_second": 75577.2315482222,
+      "required_pages_per_second_upper_estimate": 0.006550842881857765
     },
     {
       "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": 9461.881269486279,
-      "required_pages_per_second_upper_estimate": 0.0037744859061298384
+      "required_token_positions_per_second": 9447.153943527775,
+      "required_pages_per_second_upper_estimate": 0.003768610955611846
     },
     {
       "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": 18923.762538972558,
-      "required_pages_per_second_upper_estimate": 0.004377679416545577
+      "required_token_positions_per_second": 18894.30788705555,
+      "required_pages_per_second_upper_estimate": 0.004370865601208747
     },
     {
       "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": 28385.643808458837,
-      "required_pages_per_second_upper_estimate": 0.0017285367799718396
+      "required_token_positions_per_second": 28341.46183058333,
+      "required_pages_per_second_upper_estimate": 0.0017258463293241442
     },
     {
       "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": 28385.643808458837,
-      "required_pages_per_second_upper_estimate": 0.002185205436866342
+      "required_token_positions_per_second": 28341.46183058333,
+      "required_pages_per_second_upper_estimate": 0.0021818041859059433
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28385.643808458837,
-      "required_pages_per_second_upper_estimate": 0.3524546954624438
+      "required_token_positions_per_second": 28341.46183058333,
+      "required_pages_per_second_upper_estimate": 0.3519061031647979
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.14054275935933447,
-    "producer_p95_seconds": 0.36773979399731616,
-    "gpu_update_seconds": 0.9740134902897984,
-    "producer_batches_per_second": 7.115272281250985,
-    "gpu_batches_per_second": 1.026679825248077,
-    "producer_headroom": 6.9303711890235284,
+    "producer_batch_seconds": 0.3847994931095684,
+    "producer_p95_seconds": 0.3669349809970299,
+    "gpu_update_seconds": 0.9755318961763993,
+    "producer_batches_per_second": 2.5987560220492765,
+    "gpu_batches_per_second": 1.0250818081084827,
+    "producer_headroom": 2.5351693898895675,
     "sustainable": true,
-    "buffer_seconds": 60.3888363979675,
+    "buffer_seconds": 60.48297756293676,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382400.86941093445,
-    "upload_bytes_per_second": 4193017.0220391084,
+    "generation_bytes_per_second": 347946.82727739867,
+    "upload_bytes_per_second": 4149427.474250799,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.05432473894143,
+    "estimated_full_shards_per_hour": 74.66128934613675,
     "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": 1.171979762999399,
-      "bytes": 12042987,
-      "pages": 27,
-      "shards": 4
-    },
     {
       "seconds": 1.1372878729998774,
       "bytes": 6231473,
@@ -562,6 +556,12 @@
       "bytes": 1078556,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.215074095998716,
+      "bytes": 7808319,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -305,3 +305,4 @@
 {"step": 345500, "loss": 0.5766577012836933, "lr": 0.0003, "grad_norm": 0.1543777721993438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50572044161, "tokens_per_second": 148827.2971557669, "data_wait_s": 7.553000614279881e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 345600, "loss": 0.5928113032132387, "lr": 0.0003, "grad_norm": 0.15711946673021948, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50586697992, "tokens_per_second": 149801.04294844586, "data_wait_s": 1.4467001165030524e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 345700, "loss": 0.5666700676083565, "lr": 0.0003, "grad_norm": 0.14349329657627782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50601278274, "tokens_per_second": 146473.16475754857, "data_wait_s": 1.4650999219156802e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 345800, "loss": 0.5773425437510014, "lr": 0.0003, "grad_norm": 0.1719673929543946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50615996588, "tokens_per_second": 149396.85238823932, "data_wait_s": 1.3353997928788885e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:86b8e7d6f0dd6f003af52fd38d940ddbbfc59f8cff94062fc4f2c0c06fb8dd7e
-size 655089
+oid sha256:b47e5976d58cfedc0dda5975e1bbdd42aa0ebe5e92be6df79584e1682d4fa78b
+size 657207