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": 75577.88426301107,
-      "required_pages_per_second_upper_estimate": 0.006550899457521374
+      "required_token_positions_per_second": 75727.39184426649,
+      "required_pages_per_second_upper_estimate": 0.006563858395741093
     },
     {
       "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": 9447.235532876384,
-      "required_pages_per_second_upper_estimate": 0.0037686435028228757
+      "required_token_positions_per_second": 9465.92398053331,
+      "required_pages_per_second_upper_estimate": 0.003776098604010416
     },
     {
       "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": 18894.47106575277,
-      "required_pages_per_second_upper_estimate": 0.004370903349728501
+      "required_token_positions_per_second": 18931.84796106662,
+      "required_pages_per_second_upper_estimate": 0.00437954983664847
     },
     {
       "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": 28341.706598629153,
-      "required_pages_per_second_upper_estimate": 0.0017258612344139364
+      "required_token_positions_per_second": 28397.77194159993,
+      "required_pages_per_second_upper_estimate": 0.0017292753196487354
     },
     {
       "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": 28341.706598629153,
-      "required_pages_per_second_upper_estimate": 0.002181823028827672
+      "required_token_positions_per_second": 28397.77194159993,
+      "required_pages_per_second_upper_estimate": 0.002186139094128365
     },
     {
       "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": 28341.706598629153,
-      "required_pages_per_second_upper_estimate": 0.3519091423647411
+      "required_token_positions_per_second": 28397.77194159993,
+      "required_pages_per_second_upper_estimate": 0.3526052862858057
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10820478787501031,
-    "producer_p95_seconds": 0.1811451269995814,
-    "gpu_update_seconds": 0.9755234711708325,
-    "producer_batches_per_second": 9.24173522852909,
-    "gpu_batches_per_second": 1.0250906611193993,
-    "producer_headroom": 9.015529629776463,
+    "producer_batch_seconds": 0.0498234212187505,
+    "producer_p95_seconds": 0.1714220789999672,
+    "gpu_update_seconds": 0.9735975081727595,
+    "producer_batches_per_second": 20.070881837067844,
+    "gpu_batches_per_second": 1.0271184874710624,
+    "producer_headroom": 19.54096054339915,
     "sustainable": true,
-    "buffer_seconds": 60.48245521259162,
+    "buffer_seconds": 60.36304550671109,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 171,
+    "upload_pages_per_commit": 172,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 651361.9962313853,
-    "upload_bytes_per_second": 3189031.2867444735,
+    "generation_bytes_per_second": 589976.1543646867,
+    "upload_bytes_per_second": 2575337.5384665215,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.88356073000989,
+    "estimated_full_shards_per_hour": 63.297574392344714,
     "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": [
@@ -190,6 +190,12 @@
       "bytes": 7278265,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.9598245239999414,
+      "bytes": 1062466,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -30,3 +30,4 @@
 {"step": 158000, "loss": 0.5854500038549304, "lr": 0.0003, "grad_norm": 0.13975316095131562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23124097242, "tokens_per_second": 149803.65281622653, "data_wait_s": 1.4758000361325685e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 158100, "loss": 0.5768682304769754, "lr": 0.0003, "grad_norm": 0.14110618002533135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23138787938, "tokens_per_second": 149399.6597650285, "data_wait_s": 1.816199983295519e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 158200, "loss": 0.6060286071151495, "lr": 0.0003, "grad_norm": 0.15385834677023372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23153325704, "tokens_per_second": 147717.26208064603, "data_wait_s": 1.550700017105555e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 158300, "loss": 0.5833310037851334, "lr": 0.0003, "grad_norm": 0.14498027515815276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23167980281, "tokens_per_second": 148323.92403910952, "data_wait_s": 1.4323999948828714e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e50fa2a7ef1f56686fd5652e348de00a9c5238bf3a7b48d3cf9d4c6d763961a1
-size 67941
+oid sha256:b8250e5254c2afddc2ebc5c99e3354272c88b104c66d9843b2ea5c5c4df02809
+size 70059