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": 75917.63148667454,
-      "required_pages_per_second_upper_estimate": 0.006580347885787057
+      "required_token_positions_per_second": 76119.77437204753,
+      "required_pages_per_second_upper_estimate": 0.006597869118764729
     },
     {
       "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": 9489.703935834317,
-      "required_pages_per_second_upper_estimate": 0.003785584783722003
+      "required_token_positions_per_second": 9514.97179650594,
+      "required_pages_per_second_upper_estimate": 0.003795664511132097
     },
     {
       "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": 18979.407871668634,
-      "required_pages_per_second_upper_estimate": 0.004390551984940407
+      "required_token_positions_per_second": 19029.94359301188,
+      "required_pages_per_second_upper_estimate": 0.0044022425346748485
     },
     {
       "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": 28469.11180750295,
-      "required_pages_per_second_upper_estimate": 0.0017336195431908797
+      "required_token_positions_per_second": 28544.915389517824,
+      "required_pages_per_second_upper_estimate": 0.0017382355836249273
     },
     {
       "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": 28469.11180750295,
-      "required_pages_per_second_upper_estimate": 0.002191631034486968
+      "required_token_positions_per_second": 28544.915389517824,
+      "required_pages_per_second_upper_estimate": 0.0021974666040682177
     },
     {
       "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": 28469.11180750295,
-      "required_pages_per_second_upper_estimate": 0.3534910886611489
+      "required_token_positions_per_second": 28544.915389517824,
+      "required_pages_per_second_upper_estimate": 0.35443231545150455
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.042958967867161135,
-    "producer_p95_seconds": 0.12086820500007889,
-    "gpu_update_seconds": 0.9711578003186404,
-    "producer_batches_per_second": 23.278026676344428,
-    "gpu_batches_per_second": 1.0296987777598001,
-    "producer_headroom": 22.606637182757286,
+    "producer_batch_seconds": 0.05155565389844341,
+    "producer_p95_seconds": 0.1312729589999435,
+    "gpu_update_seconds": 0.9685788037106186,
+    "producer_batches_per_second": 19.396514725035665,
+    "gpu_batches_per_second": 1.032440516113926,
+    "producer_headroom": 18.78705302853044,
     "sustainable": true,
-    "buffer_seconds": 60.21178361975571,
+    "buffer_seconds": 60.05188583005835,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -23,3 +23,4 @@
 {"step": 357300, "loss": 0.5804549641907215, "lr": 0.0003, "grad_norm": 0.16365984136978362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52298628386, "tokens_per_second": 151215.89184892506, "data_wait_s": 1.5361000350821996e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 357400, "loss": 0.6018977463245392, "lr": 0.0003, "grad_norm": 0.1774761756098266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52313238998, "tokens_per_second": 150203.83542135483, "data_wait_s": 1.2816999969800236e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 357500, "loss": 0.6066079773008823, "lr": 0.0003, "grad_norm": 0.1540605306034135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52327985290, "tokens_per_second": 151886.11674555767, "data_wait_s": 1.3465999927575467e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 357600, "loss": 0.5449966806918383, "lr": 0.0003, "grad_norm": 0.16280325971500753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52342744877, "tokens_per_second": 150667.2497874788, "data_wait_s": 1.5768000139360083e-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 sha256:7e8449bc0210c942d681b0a2157c9b707cc0674aa0c923dabad2cc1b239983d5
-size 53115
+oid sha256:43a3cd471ec875a0fcdbcaa43bb5a887ed8e25bd958533fffa21e37fd835e1d2
+size 55233