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": 76465.54386158458,
-      "required_pages_per_second_upper_estimate": 0.005290063748850506
+      "required_token_positions_per_second": 76338.54112201113,
+      "required_pages_per_second_upper_estimate": 0.00528127740463986
     },
     {
       "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": 9558.192982698072,
-      "required_pages_per_second_upper_estimate": 0.004001815804089186
+      "required_token_positions_per_second": 9542.317640251391,
+      "required_pages_per_second_upper_estimate": 0.003995169129721608
     },
     {
       "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": 19116.385965396144,
-      "required_pages_per_second_upper_estimate": 0.003981439063592634
+      "required_token_positions_per_second": 19084.635280502782,
+      "required_pages_per_second_upper_estimate": 0.003974826233251214
     },
     {
       "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": 28674.578948094215,
-      "required_pages_per_second_upper_estimate": 0.0017189972243913442
+      "required_token_positions_per_second": 28626.95292075417,
+      "required_pages_per_second_upper_estimate": 0.0017161421167730415
     },
     {
       "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": 28674.578948094215,
-      "required_pages_per_second_upper_estimate": 0.0022293223870709544
+      "required_token_positions_per_second": 28626.95292075417,
+      "required_pages_per_second_upper_estimate": 0.002225619672930021
     },
     {
       "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": 28674.578948094215,
-      "required_pages_per_second_upper_estimate": 0.36780671038203994
+      "required_token_positions_per_second": 28626.95292075417,
+      "required_pages_per_second_upper_estimate": 0.3671958148401659
     }
   ],
   "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": 0.06432779949999201,
-    "producer_p95_seconds": 0.22280672000124468,
-    "gpu_update_seconds": 0.9641989878926385,
-    "producer_batches_per_second": 15.545378635252776,
-    "gpu_batches_per_second": 1.0371303149628983,
-    "producer_headroom": 14.988838346518571,
+    "producer_batch_seconds": 0.11328907967195789,
+    "producer_p95_seconds": 0.2320734029999585,
+    "gpu_update_seconds": 0.9658031043868297,
+    "producer_batches_per_second": 8.826976111869033,
+    "gpu_batches_per_second": 1.0354077300619997,
+    "producer_headroom": 8.5251209311915,
     "sustainable": true,
-    "buffer_seconds": 60.744536237236225,
+    "buffer_seconds": 60.84559557637027,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309629.8204651261,
-    "upload_bytes_per_second": 4833780.099890554,
+    "generation_bytes_per_second": 311099.13373310445,
+    "upload_bytes_per_second": 4885801.773037148,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.43935165849054,
+    "estimated_full_shards_per_hour": 66.75463208193636,
     "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": 0.9886994910002613,
-      "bytes": 1117526,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.8489439929999207,
       "bytes": 17752294,
@@ -562,6 +556,12 @@
       "bytes": 1035853,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.3170902919991931,
+      "bytes": 7309091,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -149,3 +149,4 @@
 {"step": 734900, "loss": 0.5318667888641357, "lr": 5e-05, "grad_norm": 0.22829058763692353, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107569410507, "tokens_per_second": 141545.6795911921, "data_wait_s": 2.043899985437747e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 735000, "loss": 0.5981597900390625, "lr": 5e-05, "grad_norm": 0.22974377017570943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107583954757, "tokens_per_second": 150092.4646918423, "data_wait_s": 1.6990999938570894e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 735100, "loss": 0.5108974575996399, "lr": 5e-05, "grad_norm": 0.2433496070831332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107598529665, "tokens_per_second": 138880.812832623, "data_wait_s": 1.4921999536454678e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 735200, "loss": 0.5171884894371033, "lr": 5e-05, "grad_norm": 0.2305606743511926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107613043877, "tokens_per_second": 145677.1785847433, "data_wait_s": 1.855300070019439e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c15c2dec2288d26a0ed195d4df67e838f3f35ce185a0159eac59ac36d6707aa
-size 322332
+oid sha256:c3467362f564d354579a12a87bced5013e6a39d6bdd9c5df457fac05e1a47178
+size 324450