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": 74380.76299345151,
-      "required_pages_per_second_upper_estimate": 0.00644713601995207
+      "required_token_positions_per_second": 74689.75830527968,
+      "required_pages_per_second_upper_estimate": 0.006473918950439878
     },
     {
       "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": 9297.59537418144,
-      "required_pages_per_second_upper_estimate": 0.0037089498061997125
+      "required_token_positions_per_second": 9336.21978815996,
+      "required_pages_per_second_upper_estimate": 0.003724357662422196
     },
     {
       "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": 18595.19074836288,
-      "required_pages_per_second_upper_estimate": 0.00430167011545404
+      "required_token_positions_per_second": 18672.43957631992,
+      "required_pages_per_second_upper_estimate": 0.004319540272268964
     },
     {
       "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": 27892.78612254432,
-      "required_pages_per_second_upper_estimate": 0.0016985243327240806
+      "required_token_positions_per_second": 28008.65936447988,
+      "required_pages_per_second_upper_estimate": 0.0017055804052180398
     },
     {
       "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": 27892.78612254432,
-      "required_pages_per_second_upper_estimate": 0.002147263887887248
+      "required_token_positions_per_second": 28008.65936447988,
+      "required_pages_per_second_upper_estimate": 0.0021561841308091107
     },
     {
       "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": 27892.78612254432,
-      "required_pages_per_second_upper_estimate": 0.3463350524919518
+      "required_token_positions_per_second": 28008.65936447988,
+      "required_pages_per_second_upper_estimate": 0.3477738103539973
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09927552996877154,
-    "producer_p95_seconds": 0.17008235100001912,
-    "gpu_update_seconds": 0.991224034721061,
-    "producer_batches_per_second": 10.072975690127905,
-    "gpu_batches_per_second": 1.0088536647332291,
-    "producer_headroom": 9.984575605215745,
+    "producer_batch_seconds": 0.06483073253905403,
+    "producer_p95_seconds": 0.16410649899989949,
+    "gpu_update_seconds": 0.9871232907014014,
+    "producer_batches_per_second": 15.424783290819676,
+    "gpu_batches_per_second": 1.0130446818749956,
+    "producer_headroom": 15.22616284038991,
     "sustainable": true,
-    "buffer_seconds": 60.46466611798472,
+    "buffer_seconds": 60.21452073278548,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 158,
+    "upload_pages_per_commit": 173,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 583639.5373829543,
-    "upload_bytes_per_second": 5664216.388537955,
+    "generation_bytes_per_second": 577616.0848983192,
+    "upload_bytes_per_second": 5703529.348955929,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.617729144651754,
+    "estimated_full_shards_per_hour": 61.971482802449145,
     "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": [
@@ -358,6 +358,12 @@
       "bytes": 6003887,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1132295069996871,
+      "bytes": 7572233,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -33,3 +33,4 @@
 {"step": 85200, "loss": 0.602685734629631, "lr": 0.0003, "grad_norm": 0.13874306371124315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12468770344, "tokens_per_second": 148680.2669662592, "data_wait_s": 1.0623999969539e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 85300, "loss": 0.6185211539268494, "lr": 0.0003, "grad_norm": 0.1325468472825569, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12483365755, "tokens_per_second": 147805.61911232604, "data_wait_s": 2.027199980147998e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 85400, "loss": 0.6123001296073198, "lr": 0.0003, "grad_norm": 0.13985116672214076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12497887970, "tokens_per_second": 146906.56294109672, "data_wait_s": 1.112200061470503e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 85500, "loss": 0.654580932110548, "lr": 0.0003, "grad_norm": 0.14186616915239358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12512511304, "tokens_per_second": 147532.8353491967, "data_wait_s": 1.2529000741778873e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4d96a87f47341e4fe0e4d0c13af59847aafc54d6f29ced4f2b07003152c56138
-size 75078
+oid sha256:8bbbb8aaa775c6823e9195caa1c74c316668b581ae7e441a8d2b60f48c212bbd
+size 77196