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": 75481.50050459306,
-      "required_pages_per_second_upper_estimate": 0.005221985346118669
+      "required_token_positions_per_second": 76431.8966259894,
+      "required_pages_per_second_upper_estimate": 0.005287735955019689
     },
     {
       "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": 9435.187563074132,
-      "required_pages_per_second_upper_estimate": 0.003950316003537894
+      "required_token_positions_per_second": 9553.987078248674,
+      "required_pages_per_second_upper_estimate": 0.004000054879725495
     },
     {
       "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": 18870.375126148265,
-      "required_pages_per_second_upper_estimate": 0.003930201493519413
+      "required_token_positions_per_second": 19107.97415649735,
+      "required_pages_per_second_upper_estimate": 0.003979687105633333
     },
     {
       "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": 28305.5626892224,
-      "required_pages_per_second_upper_estimate": 0.0016968752631271809
+      "required_token_positions_per_second": 28661.96123474602,
+      "required_pages_per_second_upper_estimate": 0.0017182408117422526
     },
     {
       "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": 28305.5626892224,
-      "required_pages_per_second_upper_estimate": 0.0022006329960746556
+      "required_token_positions_per_second": 28661.96123474602,
+      "required_pages_per_second_upper_estimate": 0.0022283414153575877
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28305.5626892224,
-      "required_pages_per_second_upper_estimate": 0.3630733660320211
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28661.96123474602,
+      "required_pages_per_second_upper_estimate": 0.36764486390305434
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07660711349220861,
-    "producer_p95_seconds": 0.16033565599991562,
-    "gpu_update_seconds": 0.976769135577977,
-    "producer_batches_per_second": 13.053618057306203,
-    "gpu_batches_per_second": 1.0237833727293981,
-    "producer_headroom": 12.75037122600005,
+    "producer_batch_seconds": 0.07647387233588177,
+    "producer_p95_seconds": 0.15999912800089078,
+    "gpu_update_seconds": 0.9646234524413205,
+    "producer_batches_per_second": 13.076361500407467,
+    "gpu_batches_per_second": 1.036673945122469,
+    "producer_headroom": 12.613764975893817,
     "sustainable": true,
-    "buffer_seconds": 60.55968640583457,
+    "buffer_seconds": 60.77127750380319,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 457183.4905391766,
-    "upload_bytes_per_second": 2341221.510390042,
+    "generation_bytes_per_second": 447011.7670146253,
+    "upload_bytes_per_second": 2849908.97094927,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.10093438273881,
+    "estimated_full_shards_per_hour": 95.91831929997511,
     "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": [
@@ -196,6 +196,12 @@
       "bytes": 2430008,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.2644712469991646,
+      "bytes": 6316580,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -43,3 +43,4 @@
 {"step": 804300, "loss": 0.5251337289810181, "lr": 3e-05, "grad_norm": 0.27362205489081703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117726234602, "tokens_per_second": 150082.42088628974, "data_wait_s": 1.4876999557600357e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 804400, "loss": 0.5062570571899414, "lr": 3e-05, "grad_norm": 0.26506182470806827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117740745713, "tokens_per_second": 149590.36487055628, "data_wait_s": 1.1994000487902667e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 804500, "loss": 0.5049862861633301, "lr": 3e-05, "grad_norm": 0.27502840904743453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117755350888, "tokens_per_second": 150866.8304567976, "data_wait_s": 1.8255999748362228e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 804600, "loss": 0.534039318561554, "lr": 3e-05, "grad_norm": 0.29952242586593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117770096878, "tokens_per_second": 153258.3452685798, "data_wait_s": 1.2473999959183857e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4e748ecf58ab630c080b21e7b21f7df41d8fae9bc2f85fe0a73d0c8d8f0b3e5d
-size 95475
+oid sha256:7bf65b38f903da7be6933023e2371eca1db0206d8d6d3a53222414da4a3f9ba5
+size 97593