Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76829.37827852312,
-      "required_pages_per_second_upper_estimate": 0.005315234658026471
+      "required_token_positions_per_second": 77081.22120721787,
+      "required_pages_per_second_upper_estimate": 0.005332657736189678
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9603.67228481539,
-      "required_pages_per_second_upper_estimate": 0.004020857038169882
+      "required_token_positions_per_second": 9635.152650902233,
+      "required_pages_per_second_upper_estimate": 0.004034037210065646
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19207.34456963078,
-      "required_pages_per_second_upper_estimate": 0.004000383342115006
+      "required_token_positions_per_second": 19270.305301804467,
+      "required_pages_per_second_upper_estimate": 0.004013496402240622
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28811.01685444617,
-      "required_pages_per_second_upper_estimate": 0.0017271764685485233
+      "required_token_positions_per_second": 28905.4579527067,
+      "required_pages_per_second_upper_estimate": 0.0017328380681860347
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28811.01685444617,
-      "required_pages_per_second_upper_estimate": 0.0022399298341628923
+      "required_token_positions_per_second": 28905.4579527067,
+      "required_pages_per_second_upper_estimate": 0.0022472722141501584
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28811.01685444617,
-      "required_pages_per_second_upper_estimate": 0.3695567893491126
+      "required_token_positions_per_second": 28905.4579527067,
+      "required_pages_per_second_upper_estimate": 0.3707681783546478
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12103988675801247,
-    "producer_p95_seconds": 0.35994908699649386,
-    "gpu_update_seconds": 0.9596329119405346,
-    "producer_batches_per_second": 8.261739388431831,
-    "gpu_batches_per_second": 1.0420651350711143,
-    "producer_headroom": 7.9282370270146485,
+    "producer_batch_seconds": 0.3801006732966812,
+    "producer_p95_seconds": 0.36263490399869625,
+    "gpu_update_seconds": 0.9564975599153601,
+    "producer_batches_per_second": 2.630881948529112,
+    "gpu_batches_per_second": 1.0454809734051902,
+    "producer_headroom": 2.5164321641934637,
     "sustainable": true,
-    "buffer_seconds": 60.45687345225368,
+    "buffer_seconds": 60.259346274667685,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 378218.36508235434,
-    "upload_bytes_per_second": 4335275.580641897,
+    "generation_bytes_per_second": 368795.892843083,
+    "upload_bytes_per_second": 4411231.053624848,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.15685667374585,
+    "estimated_full_shards_per_hour": 79.13501347512596,
     "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": 1.379372390001663,
-      "bytes": 7458410,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.3072389689987176,
       "bytes": 4796842,
@@ -562,6 +556,12 @@
       "bytes": 1112974,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.6805495459993836,
+      "bytes": 16247090,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -431,3 +431,4 @@
 {"step": 647500, "loss": 0.5511454343795776, "lr": 0.0001, "grad_norm": 0.17809451745051597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94777372465, "tokens_per_second": 152605.32640742042, "data_wait_s": 1.2502001482062042e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 647600, "loss": 0.5761297941207886, "lr": 0.0001, "grad_norm": 0.1767136607816377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94792018838, "tokens_per_second": 151146.24248764437, "data_wait_s": 1.1874995834659785e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 647700, "loss": 0.5617843270301819, "lr": 0.0001, "grad_norm": 0.18590474809886204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94806559444, "tokens_per_second": 150624.9594513292, "data_wait_s": 1.0046001989394426e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 647800, "loss": 0.5408211946487427, "lr": 0.0001, "grad_norm": 0.17289111194633108, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94821128922, "tokens_per_second": 149982.78966636682, "data_wait_s": 1.764699845807627e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8f8aa38f53e7be39ab61af3298d8af9acf17d1c7be0db94a35458e84ee7e86dd
-size 924306
+oid sha256:429f315e1386761c17f9f7b68da0b74eebe8ce619b090e84689f259cf9801ff1
+size 926424