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": 75297.39325350254,
-      "required_pages_per_second_upper_estimate": 0.006526587207715139
+      "required_token_positions_per_second": 75561.19670235415,
+      "required_pages_per_second_upper_estimate": 0.006549453022057868
     },
     {
       "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": 9412.174156687817,
-      "required_pages_per_second_upper_estimate": 0.003754656995646967
+      "required_token_positions_per_second": 9445.149587794269,
+      "required_pages_per_second_upper_estimate": 0.0037678113881419615
     },
     {
       "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": 18824.348313375634,
-      "required_pages_per_second_upper_estimate": 0.0043546816850843426
+      "required_token_positions_per_second": 18890.299175588538,
+      "required_pages_per_second_upper_estimate": 0.0043699382563617654
     },
     {
       "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": 28236.522470063454,
-      "required_pages_per_second_upper_estimate": 0.001719456072842758
+      "required_token_positions_per_second": 28335.448763382807,
+      "required_pages_per_second_upper_estimate": 0.001725480165079758
     },
     {
       "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": 28236.522470063454,
-      "required_pages_per_second_upper_estimate": 0.002173725663442381
+      "required_token_positions_per_second": 28335.448763382807,
+      "required_pages_per_second_upper_estimate": 0.0021813412833476117
     },
     {
       "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": 28236.522470063454,
-      "required_pages_per_second_upper_estimate": 0.3506031075165881
+      "required_token_positions_per_second": 28335.448763382807,
+      "required_pages_per_second_upper_estimate": 0.3518314409946088
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.10836200707828425,
-    "producer_p95_seconds": 0.2177893280004355,
-    "gpu_update_seconds": 0.9791574025913634,
-    "producer_batches_per_second": 9.228326670597449,
-    "gpu_batches_per_second": 1.0212862583211608,
-    "producer_headroom": 9.035984373046801,
+    "producer_batch_seconds": 0.42135872071875724,
+    "producer_p95_seconds": 0.2143757829999231,
+    "gpu_update_seconds": 0.9757389138558067,
+    "producer_batches_per_second": 2.373274720158139,
+    "gpu_batches_per_second": 1.0248643215922602,
+    "producer_headroom": 2.315696497728546,
     "sustainable": true,
-    "buffer_seconds": 60.707758960664535,
+    "buffer_seconds": 60.495812659060014,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319411.38920092944,
+    "generation_bytes_per_second": 306879.17931769404,
     "upload_bytes_per_second": 5770306.727559882,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.53824860592758,
+    "estimated_full_shards_per_hour": 65.8491280999004,
     "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": [

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

@@ -104,3 +104,4 @@
 {"step": 365400, "loss": 0.5744970254600048, "lr": 0.0003, "grad_norm": 0.1481814624720664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53484263595, "tokens_per_second": 149064.91951471538, "data_wait_s": 1.444200097466819e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 365500, "loss": 0.5785230807960033, "lr": 0.0003, "grad_norm": 0.16260728728899426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53498888747, "tokens_per_second": 149437.7973033231, "data_wait_s": 1.3671999113284983e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 365600, "loss": 0.5742380302399397, "lr": 0.0003, "grad_norm": 0.15051426151775119, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53513487408, "tokens_per_second": 149022.14693153457, "data_wait_s": 1.3954000678495504e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 365700, "loss": 0.5581132955849171, "lr": 0.0003, "grad_norm": 0.13958854611713198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53528065903, "tokens_per_second": 148731.41572040573, "data_wait_s": 1.3942999430582859e-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:fabcb4109dd772d449ffed7c2d216bf468f2cb2851a7ac26f72b314b49a0a7d9
-size 226239
+oid sha256:95989256df0bb014a80ca451e4ade79c33ca008514d7448d91f879d57bcc89e5
+size 228357