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": 75432.70212736138,
-      "required_pages_per_second_upper_estimate": 0.006538315437964018
+      "required_token_positions_per_second": 75807.86542406958,
+      "required_pages_per_second_upper_estimate": 0.0065708336416800016
     },
     {
       "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": 9429.087765920172,
-      "required_pages_per_second_upper_estimate": 0.003761404087250747
+      "required_token_positions_per_second": 9475.983178008697,
+      "required_pages_per_second_upper_estimate": 0.0037801113682817524
     },
     {
       "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": 18858.175531840345,
-      "required_pages_per_second_upper_estimate": 0.004362507016737442
+      "required_token_positions_per_second": 18951.966356017394,
+      "required_pages_per_second_upper_estimate": 0.004384203873248674
     },
     {
       "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": 28287.26329776052,
-      "required_pages_per_second_upper_estimate": 0.0017225459230329606
+      "required_token_positions_per_second": 28427.94953402609,
+      "required_pages_per_second_upper_estimate": 0.0017311129766978974
     },
     {
       "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": 28287.26329776052,
-      "required_pages_per_second_upper_estimate": 0.002177631832818102
+      "required_token_positions_per_second": 28427.94953402609,
+      "required_pages_per_second_upper_estimate": 0.0021884622487302534
     },
     {
       "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": 28287.26329776052,
-      "required_pages_per_second_upper_estimate": 0.3512331387779595
+      "required_token_positions_per_second": 28427.94953402609,
+      "required_pages_per_second_upper_estimate": 0.3529799909858337
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.09944093006259891,
-    "producer_p95_seconds": 0.3754678839977714,
-    "gpu_update_seconds": 0.9774010199915264,
-    "producer_batches_per_second": 10.05622131018376,
-    "gpu_batches_per_second": 1.023121502378491,
-    "producer_headroom": 9.828960965834131,
+    "producer_batch_seconds": 0.14498561267947707,
+    "producer_p95_seconds": 0.3832673409997369,
+    "gpu_update_seconds": 0.9725639890737617,
+    "producer_batches_per_second": 6.8972360879056485,
+    "gpu_batches_per_second": 1.0282099802526798,
+    "producer_headroom": 6.708003443237024,
     "sustainable": true,
-    "buffer_seconds": 60.59886323947464,
+    "buffer_seconds": 60.298967322573226,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 475226.8614860126,
-    "upload_bytes_per_second": 5697863.411029212,
+    "generation_bytes_per_second": 505457.416217969,
+    "upload_bytes_per_second": 5715730.477866142,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.97262175975116,
+    "estimated_full_shards_per_hour": 108.45939507393172,
     "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": 2.506807755002228,
-      "bytes": 10223439,
-      "pages": 45,
-      "shards": 3
-    },
     {
       "seconds": 0.890203859002213,
       "bytes": 916282,
@@ -562,6 +556,12 @@
       "bytes": 6081727,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9660166879984899,
+      "bytes": 2720285,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -401,3 +401,4 @@
 {"step": 395100, "loss": 0.5813555866479874, "lr": 0.0003, "grad_norm": 0.15048847348970862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57831475720, "tokens_per_second": 132896.38789515386, "data_wait_s": 1.3146003766451031e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 395200, "loss": 0.5724810808897018, "lr": 0.0003, "grad_norm": 0.1682321396114497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57846131043, "tokens_per_second": 149427.95350579947, "data_wait_s": 9.98400355456397e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 395300, "loss": 0.5941524654626846, "lr": 0.0003, "grad_norm": 0.1554837989482024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57860678821, "tokens_per_second": 148201.14242640784, "data_wait_s": 9.780000254977494e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 395400, "loss": 0.5709842573851347, "lr": 0.0003, "grad_norm": 0.1689879688406585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57875232223, "tokens_per_second": 146992.06443223663, "data_wait_s": 1.1618998541962355e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e3124d3af164a64b2520b166102c9f38f77a787454ed2041f1dbc8616010a24b
-size 859983
+oid sha256:d42018e8f887ba5a7e78ffa9fef1c202828f489a07edd0b3a7c5f981cb31e75d
+size 862101