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": 76002.29965686507,
-      "required_pages_per_second_upper_estimate": 0.006587686708189657
+      "required_token_positions_per_second": 75252.69372293638,
+      "required_pages_per_second_upper_estimate": 0.0065227127656955885
     },
     {
       "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": 9500.287457108134,
-      "required_pages_per_second_upper_estimate": 0.0037898067085958727
+      "required_token_positions_per_second": 9406.586715367048,
+      "required_pages_per_second_upper_estimate": 0.00375242808176442
     },
     {
       "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": 19000.574914216268,
-      "required_pages_per_second_upper_estimate": 0.004395448607706577
+      "required_token_positions_per_second": 18813.173430734096,
+      "required_pages_per_second_upper_estimate": 0.00435209657265644
     },
     {
       "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": 28500.8623713244,
-      "required_pages_per_second_upper_estimate": 0.0017355529859452718
+      "required_token_positions_per_second": 28219.760146101144,
+      "required_pages_per_second_upper_estimate": 0.0017184353352585704
     },
     {
       "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": 28500.8623713244,
-      "required_pages_per_second_upper_estimate": 0.002194075280781129
+      "required_token_positions_per_second": 28219.760146101144,
+      "required_pages_per_second_upper_estimate": 0.0021724352533426796
     },
     {
       "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": 28500.8623713244,
-      "required_pages_per_second_upper_estimate": 0.3538853244015099
+      "required_token_positions_per_second": 28219.760146101144,
+      "required_pages_per_second_upper_estimate": 0.3503949755528657
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.040883192093687626,
-    "producer_p95_seconds": 0.1355673489997571,
-    "gpu_update_seconds": 0.9700759099772892,
-    "producer_batches_per_second": 24.459929589362964,
-    "gpu_batches_per_second": 1.0308471633146847,
-    "producer_headroom": 23.727988454381702,
+    "producer_batch_seconds": 0.038324903750137196,
+    "producer_p95_seconds": 0.1252285739992658,
+    "gpu_update_seconds": 0.9797390146783321,
+    "producer_batches_per_second": 26.092694361859166,
+    "gpu_batches_per_second": 1.0206799821361814,
+    "producer_headroom": 25.56403066439077,
     "sustainable": true,
-    "buffer_seconds": 60.14470641859193,
+    "buffer_seconds": 60.74381891005659,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -44,3 +44,4 @@
 {"step": 89400, "loss": 0.5994367077946663, "lr": 0.0003, "grad_norm": 0.14024797043852974, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13083191395, "tokens_per_second": 150982.8180724597, "data_wait_s": 8.67000017024111e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 89500, "loss": 0.6129283681511879, "lr": 0.0003, "grad_norm": 0.1431993591353761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13097858409, "tokens_per_second": 150412.4676239055, "data_wait_s": 7.837999874027446e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 89600, "loss": 0.5705464258790016, "lr": 0.0003, "grad_norm": 0.13679116210377398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13112480965, "tokens_per_second": 149779.6610328704, "data_wait_s": 9.501000022282824e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 89700, "loss": 0.5660558044910431, "lr": 0.0003, "grad_norm": 0.18036271697744438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13127084061, "tokens_per_second": 149234.8550560003, "data_wait_s": 1.1087000530096702e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a83c3c715f9e8d2f7e2cabbbe06535797c9ecd4046e6aec096429edab490c2e9
-size 97593
+oid sha256:8e47ecb527b0e5bf51dbf4f1658db744248313fe58481a40b334bc96aa434310
+size 99711